Efficient imagery exploitation employing waveletbased feature indices

0Associated
Cases 
0Associated
Defendants 
0Accused
Products 
61Forward
Citations 
0
Petitions 
1
Assignment
First Claim
1. An efficient method of data mining to facilitate ready identification of desired features within imagery data dispersed among multiple spectral bands, comprising:
 (a) selecting a wavelet type for use in said method;
(b) providing means for manipulating said data, said means at least further capable of implementing the algorithm, ${\mathrm{GDFI}}_{2n}\left(i,t\right)=\frac{{h}_{o}{z}_{i}+{h}_{1}{z}_{i+t}+\dots \text{\hspace{1em}}+{h}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}}{{g}_{o}{z}_{i}+{g}_{1}{z}_{i+t}+\dots \text{\hspace{1em}}+{g}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}},$ where GDFI_{2n }(i, t) is a waveletbased generalized difference feature index, i refers to band i of a data collector, t is the lag between bands, h_{0}, h_{1 }. . . h_{2n−1 }are high frequency coefficients g_{0}, g_{1 }. . . g_{2n−1 }are low frequency coefficients, wherein, the number of said high and low frequency coefficients is determined upon establishing the order of said wavelet type, n is the number of vanishing moments of said selected wavelet, and z_{i}, z_{i+t }. . . z_{i+(2n−1)t }are data necessary to yield at least one said feature index from the spectral signature of an image;
(c) defining at least one said feature index;
(d) initiating at least one said means for manipulating data by setting the maximum wavelet order limit, setting K=0 and setting T=1, where K is the wavelet array index, and T is the lag, defined as the number of said hyperspectral bands skipped between the ones of said hyperspectral bands that are selected for processing;
(e) setting a lag limit defined as $1\le t\le \left(\mathrm{integer}\left(\frac{m1}{2n1}\right)\right),$ where m is the number of bands in the dataset;
(f) reading at least one data set comprising said selected bands into said means for manipulating;
(g) identifying and discarding said selected bands having compromised data;
(h) incrementing said K;
(i) incrementing said T by 1;
(j) computing a reduced set of differencesum band ratios;
(k) generating at least one said defined feature index;
(l) generating the cube of each said defined feature index;
(m) selecting prespecified ones of said defined feature indices;
(n) thresholding said selected defined feature indices, wherein said thresholding results in only said selected defined feature indices being used henceforth;
(o) saving said thresholded selected defined feature indices;
(p) determining if said lag limit has been met;
(q) if said lag limit has been met, determining if said wavelet order limit has been met;
(r) if said lag limit has not been met, incrementing said method starting at step (h) and iterating said method in like manner until said lag limit has been met;
(s) if said wavelet order limit has been met, stopping; and
(t) if said wavelet order limit has not been met, setting said T=1 and incrementing said method starting at step (h) and iterating said method in like manner until said wavelet order limit has been met, wherein, if both said lag limit and said wavelet order limit have been met, said method is ended, resulting in an efficient identification of said desired features in said imagery data.
1 Assignment
0 Petitions
Accused Products
Abstract
A waveletbased band differencesum ratio method reduces the computation cost of classification and feature extraction (identification) tasks. A Generalized Difference Feature Index (GDFI), computed using wavelets such as Daubechies wavelets, is employed in a method to automatically generate a large sequence of generalized band ratio images. In select embodiments of the present invention, judicious data mining of the large set of GDFI bands produces a small subset of GDFI bands suitable to identify specific Terrain Category/Classification (TERCAT) features. Other wavelets, such as Vaidyanathan, Coiflet, Beylkin, and Symmlet and the like may be employed in select embodiments. The classification and feature extraction (identification) performance of the band ratio method of the present invention is comparable to that obtained with the same or similar data sets using much more sophisticated methods such as discriminants, neural net classification, endmember Gibbsbased partitioning, and genetic algorithms.
77 Citations
View as Search Results
METHODS AND APPARATUS FOR ASSESSING MARKING OPERATIONS BASED ON ACCELERATION INFORMATION  
Patent #
US 20110060549A1
Filed 08/13/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING DEVICE WITH TRANSMITTER FOR TRIANGULATING LOCATION DURING LOCATE OPERATIONS  
Patent #
US 20110117272A1
Filed 08/19/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

LOCATING EQUIPMENT COMMUNICATIVELY COUPLED TO OR EQUIPPED WITH A MOBILE/PORTABLE DEVICE  
Patent #
US 20110191058A1
Filed 08/11/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS AND APPARATUS FOR DISPLAYING AN ELECTRONIC RENDERING OF A MARKING OPERATION BASED ON AN ELECTRONIC RECORD OF MARKING INFORMATION  
Patent #
US 20100085376A1
Filed 10/28/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS AND APPARATUS FOR DISPENSING MARKING MATERIAL IN CONNECTION WITH UNDERGROUND FACILITY MARKING OPERATIONS BASED ON ENVIRONMENTAL INFORMATION AND/OR OPERATIONAL INFORMATION  
Patent #
US 20100247754A1
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS, APPARATUS, AND SYSTEMS FOR ANALYZING USE OF A MARKING DEVICE BY A TECHNICIAN TO PERFORM AN UNDERGROUND FACILITY MARKING OPERATION  
Patent #
US 20100262470A1
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

SYSTEMS AND METHODS FOR USING MARKING INFORMATION TO ELECTRONICALLY DISPLAY DISPENSING OF MARKERS BY A MARKING SYSTEM OR MARKING TOOL  
Patent #
US 20100090858A1
Filed 12/16/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING DEVICE DOCKING STATIONS HAVING MECHANICAL DOCKING AND METHODS OF USING SAME  
Patent #
US 20100084532A1
Filed 09/30/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS AND APPARATUS FOR GENERATING ALERTS ON A MARKING DEVICE, BASED ON COMPARING ELECTRONIC MARKING INFORMATION TO FACILITIES MAP INFORMATION AND/OR OTHER IMAGE INFORMATION  
Patent #
US 20100188215A1
Filed 02/05/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS AND APPARATUS FOR OVERLAYING ELECTRONIC MARKING INFORMATION ON FACILITIES MAP INFORMATION AND/OR OTHER IMAGE INFORMATION DISPLAYED ON A MARKING DEVICE  
Patent #
US 20100198663A1
Filed 02/05/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING DEVICE DOCKING STATIONS AND METHODS OF USING SAME  
Patent #
US 20100085694A1
Filed 09/30/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING APPARATUS HAVING ENVIRONMENTAL SENSORS AND OPERATIONS SENSORS FOR UNDERGROUND FACILITY MARKING OPERATIONS, AND ASSOCIATED METHODS AND SYSTEMS  
Patent #
US 20100263591A1
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS AND APPARATUS FOR GENERATING AN ELECTRONIC RECORD OF A MARKING OPERATION BASED ON MARKING DEVICE ACTUATIONS  
Patent #
US 20100086677A1
Filed 08/11/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING APPARATUS HAVING ENHANCED FEATURES FOR UNDERGROUND FACILITY MARKING OPERATIONS, AND ASSOCIATED METHODS AND SYSTEMS  
Patent #
US 20100189887A1
Filed 02/11/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS AND APPARATUS FOR DISPLAYING AND PROCESSING FACILITIES MAP INFORMATION AND/OR OTHER IMAGE INFORMATION ON A MARKING DEVICE  
Patent #
US 20100188407A1
Filed 02/05/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING APPARATUS CONFIGURED TO DETECT OUTOFTOLERANCE CONDITIONS IN CONNECTION WITH UNDERGROUND FACILITY MARKING OPERATIONS, AND ASSOCIATED METHODS AND SYSTEMS  
Patent #
US 20100245086A1
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING SYSTEM AND METHOD  
Patent #
US 20090013928A1
Filed 09/24/2008

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING APPARATUS AND METHODS FOR CREATING AN ELECTRONIC RECORD OF MARKING APPARATUS OPERATIONS  
Patent #
US 20090210098A1
Filed 02/02/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

ELECTRONICALLY CONTROLLED MARKING APPARATUS AND METHODS  
Patent #
US 20090204238A1
Filed 02/02/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

ELECTRONIC MANIFEST OF UNDERGROUND FACILITY LOCATE MARKS  
Patent #
US 20090202111A1
Filed 01/30/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

MARKING APPARATUS AND METHODS FOR CREATING AN ELECTRONIC RECORD OF MARKING OPERATIONS  
Patent #
US 20090208642A1
Filed 02/02/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS FOR EVALUATING OPERATION OF MARKING APPARATUS  
Patent #
US 20090201178A1
Filed 02/02/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for generating an electronic record of a marking operation based on marking device actuations  
Patent #
US 8,280,631 B2
Filed 08/11/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Locating equipment communicatively coupled to or equipped with a mobile/portable device  
Patent #
US 8,311,765 B2
Filed 08/11/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for displaying an electronic rendering of a marking operation based on an electronic record of marking information  
Patent #
US 8,361,543 B2
Filed 10/28/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Systems and methods for using marking information to electronically display dispensing of markers by a marking system or marking tool  
Patent #
US 8,374,789 B2
Filed 12/16/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking system and method  
Patent #
US 8,386,178 B2
Filed 09/24/2008

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods for evaluating operation of marking apparatus  
Patent #
US 8,401,791 B2
Filed 02/02/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Systems and methods for using location data to electronically display dispensing of markers by a marking system or marking tool  
Patent #
US 8,407,001 B2
Filed 12/16/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Electronic manifest of underground facility locate marks  
Patent #
US 8,416,995 B2
Filed 01/30/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marker detection mechanisms for use in marking devices and methods of using same  
Patent #
US 8,424,486 B2
Filed 04/24/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking apparatus having enhanced features for underground facility marking operations, and associated methods and systems  
Patent #
US 8,442,766 B2
Filed 02/11/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for generating an electronic record of a marking operation including servicerelated information and/or ticket information  
Patent #
US 8,457,893 B2
Filed 11/20/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking apparatus having operational sensors for underground facility marking operations, and associated methods and systems  
Patent #
US 8,467,969 B2
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking apparatus and marking methods using marking dispenser with machinereadable ID mechanism  
Patent #
US 8,473,209 B2
Filed 04/24/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods, apparatus, and systems for analyzing use of a marking device by a technician to perform an underground facility marking operation  
Patent #
US 8,478,525 B2
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for dispensing marking material in connection with underground facility marking operations based on environmental information and/or operational information  
Patent #
US 8,478,524 B2
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking apparatus and methods for creating an electronic record of marking apparatus operations  
Patent #
US 8,478,523 B2
Filed 02/02/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for generating alerts on a marking device, based on comparing electronic marking information to facilities map information and/or other image information  
Patent #
US 8,510,141 B2
Filed 02/05/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking device docking stations and methods of using same  
Patent #
US 8,583,264 B2
Filed 09/30/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for displaying and processing facilities map information and/or other image information on a marking device  
Patent #
US 8,589,202 B2
Filed 03/12/2013

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking device docking stations having mechanical docking and methods of using same  
Patent #
US 8,600,526 B2
Filed 09/30/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking apparatus configured to detect outoftolerance conditions in connection with underground facility marking operations, and associated methods and systems  
Patent #
US 8,612,148 B2
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking device with transmitter for triangulating location during locate operations  
Patent #
US 8,620,572 B2
Filed 08/19/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking device docking stations having security features and methods of using same  
Patent #
US 8,644,965 B2
Filed 09/30/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking apparatus and methods for creating an electronic record of marking operations  
Patent #
US 8,700,325 B2
Filed 02/02/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for assessing marking operations based on acceleration information  
Patent #
US 8,620,616 B2
Filed 08/13/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking apparatus for receiving environmental information regarding underground facility marking operations, and associated methods and systems  
Patent #
US 8,731,830 B2
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Marking apparatus having environmental sensors and operations sensors for underground facility marking operations, and associated methods and systems  
Patent #
US 8,770,140 B2
Filed 06/09/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Systems and methods for using location data to electronically display dispensing of markers by a marking system or marking tool  
Patent #
US 8,775,077 B2
Filed 11/27/2012

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

METHODS AND SYSTEMS FOR AUTOMATIC AND SEMIAUTOMATIC GEOMETRIC AND GEOGRAPHIC FEATURE EXTRACTION  
Patent #
US 20140270359A1
Filed 03/15/2013

Current Assignee
The Boeing Co.

Sponsoring Entity
The Boeing Co.

Handheld marking apparatus with location tracking system and methods for logging geographic location of same  
Patent #
US 8,903,643 B2
Filed 03/15/2013

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for generating an electronic record of environmental landmarks based on marking device actuations  
Patent #
US 8,965,700 B2
Filed 09/28/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Optical sensing methods and apparatus for detecting a color of a marking substance  
Patent #
US 9,004,004 B2
Filed 04/22/2013

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods, apparatus and systems for surface type detection in connection with locate and marking operations  
Patent #
US 9,046,413 B2
Filed 08/15/2011

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Electronically controlled marking apparatus and methods  
Patent #
US 9,086,277 B2
Filed 02/02/2009

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and marking devices with mechanisms for indicating and/or detecting marking material color  
Patent #
US 9,097,522 B2
Filed 05/25/2010

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and apparatus for overlaying electronic marking information on facilities map information and/or other image information displayed on a marking device  
Patent #
US 9,177,403 B2
Filed 03/12/2013

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Methods and systems for automatic and semiautomatic geometric and geographic feature extraction  
Patent #
US 9,292,747 B2
Filed 03/15/2013

Current Assignee
The Boeing Co.

Sponsoring Entity
The Boeing Co.

Methods and apparatus for generating output data streams relating to underground utility marking operations  
Patent #
US 9,542,863 B2
Filed 06/03/2013

Current Assignee
Certusview Technologies LLC

Sponsoring Entity
Certusview Technologies LLC

Systems and methods for object detection  
Patent #
US 10,049,434 B2
Filed 10/15/2015

Current Assignee
The Boeing Co.

Sponsoring Entity
The Boeing Co.

Parallel inverse discrete wavelet transform  
Patent #
US 6,738,523 B1
Filed 09/14/2000

Current Assignee
Eastman Kodak Company

Sponsoring Entity
Eastman Kodak Company

Adaptive wavelet coding of hyperspectral imagery  
Patent #
US 6,539,122 B1
Filed 03/30/1998

Current Assignee
General Dynamics Mission Systems Inc.

Sponsoring Entity
General Dynamics Decision Systems Inc.

Spectral mixture process conditioned by spatiallysmooth partitioning  
Patent #
US 20050047663A1
Filed 06/26/2003

Current Assignee
University of Virginia Patent Foundation

Sponsoring Entity
University of Virginia Patent Foundation

Efficient near neighbor search (ENNsearch) method for high dimensional data sets with noise  
Patent #
US 6,947,869 B2
Filed 03/29/2002

Current Assignee
NAVY UNITED STATES OF AMERICA AS REPRESENTED BY THE SECRETARY OF THE

Sponsoring Entity
NAVY UNITED STATES OF AMERICA AS REPRESENTED BY THE SECRETARY OF THE

Detecting and segmenting local deformation in a tracked video object  
Patent #
US 6,678,416 B1
Filed 02/08/2000

Current Assignee
Washington University In St Louis

Sponsoring Entity
Washington University In St Louis

System and method for quantifying an extent to which a data mining algorithm captures useful information in input data  
Patent #
US 6,684,208 B2
Filed 05/15/2001

Current Assignee
Loyola Marymount University

Sponsoring Entity
Loyola Marymount University

Video object segmentation using active contour modelling with global relaxation  
Patent #
US 6,546,117 B1
Filed 06/10/1999

Current Assignee
University of Washington

Sponsoring Entity
University of Washington

Method of searching 3Dimensional images  
Patent #
US 6,631,364 B1
Filed 12/22/1999

Current Assignee
National Research Council Canada

Sponsoring Entity
National Research Council Canada

Adaptive threshold algorithm for realtime wavelet denoising applications  
Patent #
US 6,647,252 B2
Filed 01/18/2002

Current Assignee
ARRIS Enterprises LLC

Sponsoring Entity
General Instrument Corporation

Multiscale characterization and analysis of shapes  
Patent #
US 6,393,159 B1
Filed 06/29/1999

Current Assignee
Regents of the University of California

Sponsoring Entity
Regents of the University of California

Method and apparatus for identifying sound in a composite sound signal  
Patent #
US 6,182,018 B1
Filed 08/25/1998

Current Assignee
Ford Global Technologies LLC

Sponsoring Entity
Ford Global Technologies LLC

Method and apparatus for separation of impulsive and nonimpulsive components in a signal  
Patent #
US 6,249,749 B1
Filed 08/25/1998

Current Assignee
Ford Global Technologies LLC

Sponsoring Entity
Ford Global Technologies LLC

Nonliteral pattern recognition method and system for hyperspectral imagery exploitation  
Patent #
US 6,075,891 A
Filed 07/06/1998

Current Assignee
General Dynamics Government Systems Corporation

Sponsoring Entity
General Dynamics Government Systems Corporation

Method and apparatus for multiresolution image searching  
Patent #
US 5,933,546 A
Filed 05/06/1996

Current Assignee
NEC Corporation

Sponsoring Entity


Spatial frequency feature extraction for a classification system using wavelets  
Patent #
US 5,757,309 A
Filed 12/18/1996

Current Assignee
United States Of America As Represented By The Secretary Of The Navy

Sponsoring Entity
United States Of America As Represented By The Secretary Of The Navy

Method for analyzing a signal by wavelets  
Patent #
US 5,124,930 A
Filed 02/20/1990

Current Assignee
ThomsonCSF

Sponsoring Entity
ThomsonCSF

17 Claims
 1. An efficient method of data mining to facilitate ready identification of desired features within imagery data dispersed among multiple spectral bands, comprising:
 (a) selecting a wavelet type for use in said method;
(b) providing means for manipulating said data, said means at least further capable of implementing the algorithm,${\mathrm{GDFI}}_{2n}\left(i,t\right)=\frac{{h}_{o}{z}_{i}+{h}_{1}{z}_{i+t}+\dots \text{\hspace{1em}}+{h}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}}{{g}_{o}{z}_{i}+{g}_{1}{z}_{i+t}+\dots \text{\hspace{1em}}+{g}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}},$ where GDFI_{2n }(i, t) is a waveletbased generalized difference feature index, i refers to band i of a data collector, t is the lag between bands, h_{0}, h_{1 }. . . h_{2n−1 }are high frequency coefficients g_{0}, g_{1 }. . . g_{2n−1 }are low frequency coefficients, wherein, the number of said high and low frequency coefficients is determined upon establishing the order of said wavelet type, n is the number of vanishing moments of said selected wavelet, and z_{i}, z_{i+t }. . . z_{i+(2n−1)t }are data necessary to yield at least one said feature index from the spectral signature of an image;
(c) defining at least one said feature index;
(d) initiating at least one said means for manipulating data by setting the maximum wavelet order limit, setting K=0 and setting T=1, where K is the wavelet array index, and T is the lag, defined as the number of said hyperspectral bands skipped between the ones of said hyperspectral bands that are selected for processing;
(e) setting a lag limit defined as$1\le t\le \left(\mathrm{integer}\left(\frac{m1}{2n1}\right)\right),$ where m is the number of bands in the dataset;
(f) reading at least one data set comprising said selected bands into said means for manipulating;
(g) identifying and discarding said selected bands having compromised data;
(h) incrementing said K;
(i) incrementing said T by 1;
(j) computing a reduced set of differencesum band ratios;
(k) generating at least one said defined feature index;
(l) generating the cube of each said defined feature index;
(m) selecting prespecified ones of said defined feature indices;
(n) thresholding said selected defined feature indices, wherein said thresholding results in only said selected defined feature indices being used henceforth;
(o) saving said thresholded selected defined feature indices;
(p) determining if said lag limit has been met;
(q) if said lag limit has been met, determining if said wavelet order limit has been met;
(r) if said lag limit has not been met, incrementing said method starting at step (h) and iterating said method in like manner until said lag limit has been met;
(s) if said wavelet order limit has been met, stopping; and
(t) if said wavelet order limit has not been met, setting said T=1 and incrementing said method starting at step (h) and iterating said method in like manner until said wavelet order limit has been met, wherein, if both said lag limit and said wavelet order limit have been met, said method is ended, resulting in an efficient identification of said desired features in said imagery data.  View Dependent Claims (2, 3, 4, 5, 6, 7, 8)
 (a) selecting a wavelet type for use in said method;
 9. A photograph obtained from a process employing an efficient method of data mining to facilitate ready identification of desired features within imagery data dispersed over multiple spectral bands, said method comprising:
 (a) selecting a wavelet type for use in said method;
(b) providing means for manipulating said data, said means at least further capable of implementing the algorithm,${\mathrm{GDFI}}_{2n}\left(i,t\right)=\frac{{h}_{o}{z}_{i}+{h}_{1}{z}_{i+t}+\dots +{h}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}}{{g}_{o}{z}_{i}+{g}_{1}{z}_{i+t}+\dots +{g}_{\left(2n1\right)t}{z}_{i+\left(2n1\right)t}},$ where GDFI_{2n }(i, t) is a waveletbased generalized difference feature index, i refers to band i of a data collector, t is the lag between bands, h_{0}, h_{1 }. . . h_{2n−1 }are high frequency coefficients g_{0}, g_{1 }. . . g_{2n−1 }are low frequency coefficients, wherein, the number of said high and low frequency coefficients is determined upon establishing the order of said wavelet type, n is the number of vanishing moments of said selected wavelet, and z_{i}, z_{i+t }. . . z_{i+(2n−1)t }are data necessary to yield at least one said feature index from the spectral signature of an image;
(c) defining at least one said feature index;
(d) initiating at least one said means for manipulating data by setting the maximum wavelet order limit, setting K=0 and setting T=1, where K is the wavelet array index, and T is the lag, defined as the number of said hyperspectral bands skipped between the ones of said hyperspectral bands that are selected for processing;
(e) setting a lag limit defined as$1\le t\le \left(\mathrm{integer}\left(\frac{m1}{2\text{\hspace{1em}}n1}\right)\right),$ where m is the number of bands in the dataset;
(f) reading at least one data set comprising said selected bands into said means for manipulating;
(g) identifying and discarding said selected bands having compromised data;
(h) incrementing said K;
(i) incrementing said T by 1;
(j) computing a reduced set of differencesum band ratios;
(k) generating at least one said defined feature index;
(l) generating the cube of each said defined feature index;
(m) selecting prespecified ones of said defined feature indices;
(n) thresholding said selected defined feature indices, wherein said thresholding results in only said selected defined feature indices being used henceforth;
(o) saving said thresholded selected defined feature indices;
(p) determining if said lag limit has been met;
(q) if said lag limit has been met, determining if said wavelet order limit has been met;
(r) if said lag limit has not been met, incrementing said method starting at step (h) and iterating said method in like manner until said lag limit has been met;
(s) if said wavelet order limit has been met, stopping; and
(t) if said wavelet order limit has not been met, setting said T=1 and incrementing said method starting at step (h) and iterating said method in like manner until said wavelet order limit has been met, wherein, if both said lag limit and said wavelet order limit have been met, said method is ended, resulting in an efficient identification of said desired features in said photograph.
 (a) selecting a wavelet type for use in said method;
 10. An efficient method of data mining to facilitate ready identification of desired features within a multiband data set, comprising:
 (a) selecting a wavelet type for use in said method;
(b) providing means for manipulating said data, said means at least further capable of implementing the algorithm,${\mathrm{GDFI}}_{2n}\left(i,t\right)=\frac{{h}_{o}{z}_{i}+{h}_{1}{z}_{i+t}+\dots +{h}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}}{{g}_{o}{z}_{i}+{g}_{1}{z}_{i+t}+\dots +{g}_{\left(2n1\right)t}{z}_{i+\left(2n1\right)t}},$ where GDFI_{2n }(i, t) is a waveletbased generalized difference feature index, i refers to band i of a data collector, t is the lag between bands, h_{0}, h_{1}, . . . h_{2n−1 }are high frequency coefficients g_{0}, g_{1 }. . . g_{2n−1 }are low frequency coefficients, wherein, the number of said high and low frequency coefficients is determined upon establishing the order of said wavelet type, n is the number of vanishing moments of said selected wavelet, and z_{i}, z_{i+t }. . . z_{i+(2n−1)t }are data necessary to yield at least one said feature index from the spectral signature of an image;
(c) defining at least one said feature index;
(d) initiating at least one said means for manipulating data by setting the maximum wavelet order limit, setting K=0 and setting T=1, where K is the wavelet array index, and T is the lag, defined as the number of said hyperspectral bands skipped between the ones of said hyperspectral bands that are selected for processing;
(e) setting a lag limit defined as$1\le t\le \left(\mathrm{integer}\left(\frac{m1}{2\text{\hspace{1em}}n1}\right)\right),$ where m is the number of bands in the dataset;
(f) reading at least one data set comprising said selected bands into said means for manipulating;
(g) identifying and discarding said selected bands having compromised data;
(h) incrementing said K;
(i) incrementing said T by 1;
(j) computing a reduced set of differencesum band ratios;
(k) generating at least one said defined feature index;
(l) generating the cube of each said defined feature index;
(m) selecting prespecified ones of said defined feature indices;
(n) thresholding said selected defined feature indices, wherein said thresholding results in only said selected defined feature indices being used henceforth;
(o) saving said thresholded selected defined feature indices;
(p) determining if said lag limit has been met;
(q) if said lag limit has been met, determining if said wavelet order limit has been met;
(r) if said lag limit has not been met, incrementing said method starting at step (h) and iterating said method in like manner until said lag limit has been met;
(s) if said wavelet order limit has been met, stopping; and
(t) if said wavelet order limit has not been met, setting said T=1 and incrementing said method starting at step (h) and iterating said method in like manner until said wavelet order limit has been met, wherein, if both said lag limit and said wavelet order limit have been met, said method is ended, resulting in an efficient identification of said desired features in said multiband data set.
 (a) selecting a wavelet type for use in said method;
 11. A data mining tool comprising a computer loaded with software sufficient to implement an efficient method of data mining to facilitate ready identification of desired features within a multiband data set, said method of data mining comprising:
 (a) selecting a wavelet type for use in said method;
(b) providing means for manipulating said data, said means at least further capable of implementing the algorithm,${\mathrm{GDFI}}_{2n}\left(i,t\right)=\frac{{h}_{o}{z}_{i}+{h}_{1}{z}_{i+t}+\dots +{h}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}}{{g}_{o}{z}_{i}+{g}_{1}{z}_{i+t}+\dots +{g}_{\left(2n1\right)t}{z}_{i+\left(2n1\right)t}},$ where GDFI_{2n }(i, t) is a waveletbased generalized difference feature index, i refers to band i of a data collector, t is the lag between bands, h_{0}, h_{1 }. . . h_{2n−1 }are high frequency coefficients g_{0}, g_{1 }. . . g_{2n−1 }are low frequency coefficients, wherein, the number of said high and low frequency coefficients is determined upon establishing the order of said wavelet type, n is the number of vanishing moments of said selected wavelet, and z_{i}, z_{i+t }. . . z_{i+(2n−1)t }are data necessary to yield at least one said feature index from the spectral signature of an image;
(c) defining at least one said feature index;
(d) initiating at least one said means for manipulating data by setting the maximum wavelet order limit, setting K=0 and setting T=1, where K is the wavelet array index, and T is the lag, defined as the number of said hyperspectral bands skipped between the ones of said hyperspectral bands that are selected for processing;
(e) setting a lag limit defined as$1\le t\le \left(\mathrm{integer}\left(\frac{m1}{2\text{\hspace{1em}}n1}\right)\right),$ where m is the number of bands in the dataset;
(f) reading at least one data set comprising said selected bands into said means for manipulating;
(g) identifying and discarding said selected bands having compromised data;
(h) incrementing said K;
(i) incrementing said T by 1;
(j) computing a reduced set of differencesum band ratios;
(k) generating at least one said defined feature index;
(l) generating the cube of each said defined feature index;
(m) selecting prespecified ones of said defined feature indices;
(n) thresholding said selected defined feature indices, wherein said thresholding results in only said selected defined feature indices being used henceforth;
(o) saving said thresholded selected defined feature indices;
(p) determining if said lag limit has been met;
(q) if said lag limit has been met, determining if said wavelet order limit has been met;
(r) if said lag limit has not been met, incrementing said method starting at step (h) and iterating said method in like manner until said lag limit has been met;
(s) if said wavelet order limit has been met, stopping; and
(t) if said wavelet order limit has not been met, setting said T=1 and incrementing said method starting at step (h) and iterating said method in like manner until said wavelet order limit has been met, wherein, if both said lag limit and said wavelet order limit have been met, said method is ended, resulting in an efficient identification of said desired features in said multiband data set.  View Dependent Claims (15, 16, 17)
 (a) selecting a wavelet type for use in said method;
 18. An efficient method of data mining to facilitate ready categorization of a data set dispersed over multiple bands, comprising:
 selecting at least one wavelet type for use in said method, wherein said wavelet is selected to achieve optimum computational efficiency;
providing software for at least implementing an algorithm to calculate at least one waveletbased generalized difference feature index (GDFI);
defining at least one said feature index;
providing a software routine to select and process a reduced set of bands of said imagery data;
iterating a subroutine of said routine while applying a lag limit and a wavelet order limit to establish the number of iterations, said subroutine to at least;
compute a reduced set of differencesum band ratios;
generate at least one defined feature index;
generate the cube of each said defined feature index;
select prespecified ones of said defined feature indices;
threshold said selected defined feature indices, wherein said thresholding results in only said selected defined feature indices being used henceforth; and
save said thresholded selected defined feature indices;
wherein, if both said lag limit and said wavelet order limit have been met, said method is ended, resulting in an efficient identification of said desired features in said data set.
 selecting at least one wavelet type for use in said method, wherein said wavelet is selected to achieve optimum computational efficiency;
 19. A method that samples all band ratio combinations in hyperspectral data for use with rapid combinatorial computations that integrate wavelet and waveletvariogram techniques for improved data anomaly filtering, detection and classification of imagery, comprising:
 selecting at least one wavelet type for use in said method, wherein said wavelet is selected to achieve optimum computational efficiency; and
providing efficient software that displays results in a form that facilitates classification and feature extraction tasks while employing a leastordered said wavelet that enables select features to be readily identified, wherein implementing said software yields band ratios that provide useful information in support of said classification and feature extraction tasks, and wherein said method yields select said imagery with specific features highlighted by employing at least one generalized difference feature index (GDFI) band ratio and multiplying it by an appropriate constant such as the Daubechies coefficients, said GDFI band ratio defined by;
${\mathrm{GDFI}}_{2n}\left(i,t\right)=\frac{{h}_{o}{z}_{i}+{h}_{1}{z}_{i+t}+\dots +{h}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}}{{g}_{o}{z}_{i}+{g}_{1}{z}_{i+t}+\dots +{g}_{\left(2n1\right)t}{z}_{i+\left(2n1\right)t}},$ where GDFI_{2n }(i, t) is a waveletbased generalized difference feature index, i refers to band i of a data collector, t is the lag between bands, h_{0}, h_{1 }. . . h_{2n−1 }are high frequency coefficients g_{0}, g_{1 }. . . g_{2n−1 }are low frequency coefficients, wherein, the number of said high and low frequency coefficients is determined upon establishing the order of said wavelet type, n is the number of vanishing moments of said selected wavelet, and z_{i}, z_{i+t }. . . z_{i+(2n−1)t }are data used to yield at least one said feature index; and
wherein at least one said feature becomes distinctive, i.e., appearing as a distinct color or shade lighter than the remainder of said imagery.
 selecting at least one wavelet type for use in said method, wherein said wavelet is selected to achieve optimum computational efficiency; and
 20. A data mining tool comprising a computer sufficient to implement an efficient method that samples all band ratio combinations in hyperspectral data for use with rapid combinatorial computations that integrate wavelet and waveletvariogram techniques for improved data anomaly filtering, detection and classification of imagery, said method comprising:
 selecting at least one wavelet type for use in said method, wherein said wavelet is selected to achieve optimum computational efficiency; and
providing efficient software loaded on said computer for displaying results in a form that facilitates classification and feature extraction tasks while employing a leastordered said wavelet that enables select features to be readily identified, wherein implementing said efficient software yields band ratios that provide useful information in support of said classification and feature extraction tasks, and wherein said method yields select said imagery with specific features highlighted by employing at least one generalized difference feature index (GDFI) band ratio and multiplying it by an appropriate constant such as the Daubechies coefficients, said GDFI band ratio defined by;
${\mathrm{GDFI}}_{2n}\left(i,t\right)=\frac{{h}_{o}{z}_{i}+{h}_{1}{z}_{i+t}+\dots +{h}_{\left(2n1\right)}{z}_{i+\left(2n1\right)t}}{{g}_{o}{z}_{i}+{g}_{1}{z}_{i+t}+\dots +{g}_{\left(2n1\right)t}{z}_{i+\left(2n1\right)t}},$ where GDFI_{2n }(i, t) is a waveletbased generalized difference feature index, i refers to band i of a data collector, t is the lag between bands, h_{0}, h_{1 }. . . h_{2n−1 }are high frequency coefficients g_{0}, g_{1 }. . . g_{2n−1 }are low frequency coefficients, wherein, the number of said high and low frequency coefficients is determined upon establishing the order of said wavelet type, n is the number of vanishing moments of said selected wavelet, and z_{i}, z_{i+t }. . . z_{i+(2n−1)t }are data used to yield at least one said feature index; and
wherein at least one said feature becomes distinctive, i.e., appearing as a distinct color or shade lighter than the remainder of said imagery.
 selecting at least one wavelet type for use in said method, wherein said wavelet is selected to achieve optimum computational efficiency; and
1 Specification
Under 35 U.S.C. § 119(e)(1), this application claims the benefit of prior copending U.S. Provisional Patent Application Ser. No. 60/704,416, WaveletBased Feature Indices and Their Use as a Data Mining Tool for Hyperspectral Imagery Exploitation, by Simental et al., filed Aug. 2, 2005, and incorporated herein by reference.
STATEMENT OF GOVERNMENT INTERESTUnder paragraph 1(a) of Executive Order 10096, the conditions under which this invention was made entitle the Government of the United States, as represented by the Secretary of the Army, to an undivided interest therein on any patent granted thereon by the United States. This patent has multiple assignees. This and related patents are available for licensing to qualified licensees. Please contact Jeff Walaszek at 703 428672 or Phillip Stewart at 601 6344113.
BACKGROUNDMultispectral and hyperspectral images are composed of amounts of data that are impractical to analyze manually. These data include multiple spectral bands that are not visualized or assessed readily. Conventional multispectral sensors provide only a few spectral bands of imagery, nominally covering prespecified portions of the visible to the near infrared spectra. Conventional hyperspectral sensors may cover hundreds of spectral bands spanning a prespecified portion of the electromagnetic spectrum. Thus, hyperspectral sensors may provide greater spectral discrimination than multispectral sensors and allow nonliteral processing of data to detect and classify material content as well as structure.
An image may be represented mathematically as a matrix of m rows and n columns of elements. An element of such a matrix defining a twodimensional (2D) image is termed a picture element, or pixel. An image is usable when a viewer is able to partition the image into a number of recognizable regions that correspond to known features, such as trees, lakes, and manmade objects. Once this level of imaging is attained, each distinct feature and object may be identified since each is represented by an essentially uniform field. The process that generates such uniform fields is known as segmentation.
Many techniques have been used to segment images. Segmentation may be classinterval based, edgebased, and regionbased.
For 8bit precision data, a given image may assume pixel (element) values from a minimum of zero to a maximum of 255. By mapping into one category those pixels whose intensity values are within a certain range or class interval, e.g., 020, a simple threshold method may be used to segment.
An edge may be defined by observing the difference between adjacent pixels. Edgebased segmentation generates an edge map, linking the edge pixels to form a closed contour. In conventional edgebased segmentation, welldefined mathematical formulae are used to define an edge. After edges are extracted, another set of mathematical rules may be used to join, eliminate, or both join and eliminate edges, thus generating a closed contour around a uniform region. That is, the scene itself is not used to define an edge even though, globally, an edge may be defined by the scene.
Regionbased segmentation is the antithesis of edgebased segmentation. It begins at the interior of a potentially uniform field rather than at its outer boundary. It may be initiated with any two interior adjacent pixels. One or more rules, such as a Markov Random Field (MRF) approach, are used to decide whether merging of these two candidates should occur. In general, conventional regionbased segmentation is performed on an image within but a single spectral band, follows welldefined mathematical decision rules, is computationally intensive, and thus expensive, and is not selfdetermining or selfcalibrating.
Colorbased segmentation requires input of three spectrally distinct bands or colors. A true color video image may be generated from a scene taken by three bands of blue, green and red. They may be combined into a composite image using individual filters of the same three colors. The resultant color image may be considered a segmented image because each color may represent a uniform field.
If a region or an edge may be generated from the content of the scene, it should be possible to integrate both regionbased and edgebased segmentation methods into a single, integrated process. The process by which a segment, or region, is matched with a rule set, or model, is termed identification.
Identification occurs after segmentation. It results in labeling structure using commonlyaccepted names, such as river, forest or automobile. While identification may be achieved in a number of ways, such as statistical document functions and rulebased and modelbased matching, all require extracting representative features as an intermediate step. Extracted features may be spectral reflectancebased, texturebased, and shapebased.
Statistical pattern recognition exploits standard multivariate statistical methods. Rulebased recognition schemes use conventional artificial intelligence (AI). Shape analysis employs a modelbased approach that requires extraction of features from the boundary contour or a set of depth contours. Sophisticated features that may be extracted include Fourier descriptors and moments. Structure is identified when a match is found between observed structure and a calibration sample. A set of calibration samples constitutes a calibration library. A conventional library is both feature and fullshape based.
Feature extraction utilizes a few, but effective, representative attributes to characterize structure. While it capitalizes on economy of computation, it may select incorrect features and apply incomplete information sets in the recognition process. A fullshape model assumes that structure is not contaminated by noise, obscured by ground clutter, or both. In general, this assumption does not correspond to the operation of actual sensors.
Depth contours match threedimensional (3D) structure generated from a sensor with 3D models generated from wire frames. In general, all actual images are 3D because the intensity values of the image constitute the third dimension, although all are not created equal. For example, a LADAR image has a welldefined third dimension and a general spectralbased image does not. However, most objective discrimination comes from the boundary contour, not the depth contour.
Detection, classification (segmentation), and identification techniques applied to hyperspectral imagery are inherently either fullpixel or mixedpixel techniques in which each pixel vector in the image records the spectral information. Fullpixel techniques operate on the assumption that each pixel vector measures the response of one predominate underlying material, or signal, at each site in a scene. However, the underlying assumption for mixedpixel techniques is that each pixel vector measures the response of multiple underlying materials, or signals, at each site. In actuality, an image may be represented best by a combination of the two. Although some sites represent a single material, others are mixtures of multiple materials. Rand, Robert S. and Daniel M. Keenan, A Spectral Mixture Process Conditioned by GibbsBased Partitioning, IEEE Transactions on Geoscience and Remote Sensing, Vol. 39, No. 7, pp.14211434, July 2001.
The simplest fullpixel technique involves spectral matching. Spectra of interest in an image are matched to training spectra obtained from a library or the image itself. Metrics for determining the degree of match include: Euclidian distance, derivative difference, and spectral angle. If the relative number of mixed pixels in a scene is significant, then spectral matching of this type is not employed. Class label assignments generated by spectral matching algorithms are not affected by spatial neighborhoods, however, consistency of class labels in localized spatial neighborhoods, termed “spatial localization,” is important in mapping applications.
Other fullpixel methods include various supervised and unsupervised segmentation techniques. These are based on statistical and pattern recognition methods normally applied to multispectral image processing. The training is also done using data from libraries or the scene imagery itself. Specific techniques include: statistical linear discrimination, e.g., Fisher's linear discriminant; quadratic multivariate classifiers, e.g., Mahalanobis and Bayesian maximum likelihood (ML) classifiers; and neural networks.
The quadratic methods require lowdimensional pixel vectors, and thus are preceded by a data reduction operation to reduce the number of spectral bands addressed. Effective neural networks, such as the multilayer feedforward neural network (MLFN), may be built to model quadratic and higher order decision surfaces without data reduction. Although the MLFN may be trained to identify materials perturbed by a limited amount of mixing, usually it does not include any spatial localization in the decision process.
The most common unsupervised algorithms for clustering imagery are KMEANS and ISODATA, in which the metric used in determining cluster membership is Euclidian distance. Euclidian distance does not provide an adequate response when fine structure or shapes are presented in high resolution spectra, being overly sensitive to intensity changes. Additionally, these methods do not include any spatial localization in the clustering operation.
Spectral mixture analysis (SMA) techniques used with mixedpixel approaches address some of the shortcomings of fullpixel techniques. SMA employs linear statistical modeling, signal processing techniques, or both. SMA techniques are governed by the relationship:<FORM>X<sub>s</sub>=H β<sub>s</sub>+η<sub>s</sub> (1)</FORM>where:
 X<sub>s</sub>=observed reflected energy from site s
 β<sub>s</sub>=modeling parameter vector associated with mixture proportions at site s
 η<sub>s</sub>=random variable for model error at site s
 H=the matrix containing the spectra of pure materials of interest
The matrix, H, is presumed known and fixed, although for most actual materials there exists no single fixed spectral signatures to represent the pure materials.
The basic SMA may be modified to partition the H matrix into desired and undesired signatures. Subspace projections orthogonal or oblique to the undesired signatures and noise components are computed. Orthogonal subspace projection (OSP) is applied to hyperspectral imagery to suppress undesirable signatures and detect signatures of interest. This is shown in the relationship:<FORM>H=[D,U] (2)</FORM>where:
 D=matrix of known spectra for a target of interest
 U=the matrix of undesired, but known, spectraThe matrix, U, may be unknown if D is a minor component of the scene.
The above modifications are best suited to targeting applications rather than mapping.
From the early days of multispectral remote sensing to the present, earth scientists have been thoroughly and meticulously measuring the wavelengths and intensity of visible and nearinfrared light reflected by the land surface back up into space. In some instances, they have used either a “Vegetation Index” (VI) or a “Normalized Difference Vegetation Index” (NDVI) to quantify the concentrations of green leaf vegetation around the planet. Weier, J. and D. Herring, Measuring Vegetation, NDVI and EVI, Earth Observatory, September 1999. These indices may be described mathematically as:<maths id="MATHUS00001" num="1"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><mi>VI</mi><mo>=</mo><mfrac><msub><mi>b</mi><mn>1</mn></msub><msub><mi>b</mi><mn>2</mn></msub></mfrac></mrow><mo></mo><mstyle><mtext></mtext></mstyle><mo></mo><mrow><mi>And</mi><mo></mo><mstyle><mtext> </mtext></mstyle><mo></mo><mi>NDVI</mi><mo></mo><mstyle><mtext> </mtext></mstyle><mo></mo><mi>as</mi><mo></mo><mstyle><mtext>:</mtext></mstyle></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>3</mn><mo>)</mo></mrow></mtd></mtr><mtr><mtd><mrow><mi>NVDI</mi><mo>=</mo><mfrac><mrow><mo>(</mo><mrow><msub><mi>b</mi><mn>1</mn></msub><mo></mo><msub><mi>b</mi><mn>2</mn></msub></mrow><mo>)</mo></mrow><mrow><mo>(</mo><mrow><msub><mi>b</mi><mn>1</mn></msub><mo>+</mo><msub><mi>b</mi><mn>2</mn></msub></mrow><mo>)</mo></mrow></mfrac></mrow></mtd><mtd><mrow><mo>(</mo><mn>4</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths>Where
 VI is vegetation index,
 NDVI is the Normalized Difference Vegetation Index, and
 b<sub>1 </sub>is a near infrared spectral band, and
 b<sub>2 </sub>is a visible spectral band.
VI and NDVI involve mathematical operations of a combination of two or more bands aimed at enhancing vegetation features. VI and NDVI yield estimates of vegetation health, provide a means of monitoring changes in vegetation relative to biomass and color, and serve as indicators of drought, climate change, precipitation, and the like. Kidwell, K. B., Global Vegetation Index User's Guide, U.S. Department of Commerce/National Oceanic and Atmospheric Administration, July 1997 (estimate of health); Boone, R., K. Galvin, N. Smith, and S. Lynn, Generalizing El Nino Effects upon Maasai Livestock Using Hierarchical Clusters of Vegetation Patterns, Photogrammetric Engineering & Remote Sensing, Vol. 66(6): pages 737744, June 2000 (monitoring changes); Kassa, A., Drought Risk Monitoring for the Sudan, Master of Science Dissertation, University College, London, UK, August 1999 (indicator of climate).
The usefulness of the VI and NDVI is well documented and it is clear that these techniques have contributed substantial information to vegetation studies and other investigations using remote sensing. VI and NDVI have been suggested as means for identifying features other than vegetation, but these suggestions have not been aggressively investigated. Deer, P. J., Digital Detection Techniques: Civilian and Military Applications, International Symposium on Spectral Sensing Research, Melbourne, Australia, November, 1995.
Analysis of all possible ratio combinations in hyperspectral data approaches mathematical chaos, thus, it was postulated that other differencesum band ratios may provide unexpected relationships yielding useful information about terrain features, both in multispectral and hyperspectral data. Embodiments of the present invention address reducing the number of ratio combinations to identify multiple object classes, not just vegetation.
Advances in hyperspectral sensor technology provide high quality data for the accurate generation of terrain categorization/classification (TERCAT) maps. The generation of TERCAT maps from hyperspectral imagery can be accomplished using a variety of spectral pattern analysis algorithms; however, the algorithms are sometimes complex, and the training of such algorithms can be tedious. Further, hyperspectral imagery implies large data files since contiguous spectral bands are highly correlated. The correlation further implies redundancy in classification/feature extraction computations.
The use of wavelets to generate a set of “Generalized Difference Feature Indices” (GDFI) transforms a hyperspectral image cube into a derived set of GDFI bands. Each index is a “derived band” that is a generalized ratio of the originally available bands. Thus, select embodiments of the present invention generate a set of derived bands. For example, an index may be generated with a Daubechies wavelet with two (2), four (4), eight (8) or more “vanishing moments.”
Vanishing moments, filter and smoothing coefficients, and low and high frequency coefficients are all related to the order of the wavelet, and these terms are sometimes used interchangeably, e.g., a wavelet of order four (4) may be referred to as “a wavelet with four (4) vanishing moments.” A wavelet with x vanishing moments, i.e., order x, means that the first x moments starting from zero (0) moment are equal to zero (0). This suppresses signals that are of a polynomial of degree less than or equal to x−1 .
The number of filter coefficients is chosen when the order of the wavelet is established. For example, in research into data mining for select embodiments of the present invention, the initial effort started with [Daubechies 2, lag 3], i.e., the Haar wavelet, and increased both Daubechies order and lag to perform efficient data mining. The collection of these derived bands becomes the indices for the specific feature of interest. For example if the differencesum ratios and Daubechies wavelets with one vanishing moment, i.e., two (2) filtering coefficients, and lag of 3 that “identifies” or “highlights” roads and roads may also be identified with two vanishing moments, i.e., four (4) filtering coefficients, and a lag of 5, then the indices for roads may be described as [Daubechies 2, lag 3] and [Daubechies 4, lag 5].
A commonly known special case of a GDFI is a Limited Difference Feature Index (LDFI) approach as described above for the Normalized Difference Vegetation Index (NDVI). Numerous other limited bandratio indices readily identifying individual specific scene features are LDFIs, i.e., single purpose special cases of the GDFI. Generating a set of GDFI bands is fast and simple. However, there are a large number of possible bands and only a few “generalized ratios (indices)” prove useful. Judicious data mining of the large set of GDFI bands produces a small subset of GDFI bands suitable to identify specific TERCAT features.
In select embodiments of the present invention, a waveletbased differencesum band ratio method reduces the computation cost of classification and feature extraction (identification) tasks. A Generalized Difference Feature Index (GDFI), computed using wavelets such as Daubechies wavelets, is employed in an embodiment of the present invention as a method to automatically generate a large sequence of generalized band ratio images. Other wavelets, such as Vaidyanathan, Coiflet, Beylkin, and Symmlet and the like may be employed in select embodiments of the present invention. Selection of the optimum wavelet is important for computational efficiency. Simental, E., and T. Evans, Wavelet Denoising of Hyperspectral Data, International Symposium for Spectral Sensing Research, San Diego, Calif., June 1997.
A description of a method for analyzing a signal by wavelets is provided in U. S. Pat. No. 5,124,930, Method for Analyzing a Signal by Wavelets, to Nicolas et al., Jun. 23, 1992, incorporated herein by reference.
The classification and feature extraction performance of a band ratio method of the present invention was comparable to results obtained with the same data sets using much more sophisticated methods such as discriminants and neural net classification and endmember Gibbsbased partitioning. Rand, R. S., and E. H. Bosch, The Effect of Waveletbased Dimension Reduction on Neural Network Classification and Subpixel Targeting Algorithms, SPIE Defense and Security Symposium, Orlando, Fla., April 2004 (discriminants and neural net). Rand and Keenan (2001) (endmember partitioning). The performance of an embodiment of the present invention was comparable to results obtained from a similar data set using genetic algorithms. Simental, E., D. Ragsdale, E. Bosch, R. Dodge Jr., and R. Pazak, Hyperspectral Dimension Reduction and Elevation Data for Supervised Image Classification, American Society for Photogrammetry and Remote Sensing Conference, Anchorage, Ak., May 2003.
Select embodiments of the present invention extract (identify) features from hyperspectral imagery rapidly and reliably, inexpensively permitting ready identification of prespecified features.
BRIEF DESCRIPTION OF DRAWINGSThe patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.
FIG. 1, provided for illustration purposes only, is a plot of wavelength, nanometers (ηm), versus radiance for most of the features in a hyperspectral data scene.
FIG. 1A is a color rendition of FIG. 1.
FIG. 2, provided for illustration purposes only, depicts two images generated with an embodiment of the present invention by using two different ratios from a hyperspectral data set.
FIG. 3 depicts a standard HYDICE hyperspectral composite class map of the area around Copperas Cove, Tex.
FIG. 3A is a color rendition of FIG. 3.
FIG. 4 depicts a standard HYDICE hyperspectral composite class map of the area around Yuma City, Ariz.
FIG. 4A is a color rendition of FIG. 4.
FIG. 5 depicts a first enhanced image of FIG. 3 generated with an embodiment of the present invention to emphasize vegetation, but not grass.
FIG. 5A is a color rendition of FIG. 5.
FIG. 6 depicts a second enhanced image of FIG. 3 generated with an embodiment of the present invention to emphasize buildings.
FIG. 6A is a color rendition of FIG. 6.
FIG. 7 depicts a third enhanced image of FIG. 3 generated with an embodiment of the present invention to emphasize grass.
FIG. 7A is a color rendition of FIG. 7.
FIG. 8, shown for illustrative purposes only, is a ground truth photo of a typical grassy area in Copperas Cove.
FIG. 8A is a color rendition of FIG. 8.
FIG. 9 depicts a fourth enhanced image of FIG. 3 generated with an embodiment of the present invention to emphasize dirt roads, concrete, and bright rooftops.
FIG. 9A is a color rendition of FIG. 9.
FIG. 10 depicts a class map of FIG. 3, generated from data used to create FIGS. 5, 6, 7, and 9.
FIG. 10A is a color rendition of FIG. 10.
FIG. 11 depicts a first enhanced image of FIG. 4 generated with an embodiment of the present invention to emphasize vegetation such as trees, shrubs, bushes, and grass.
FIG. 11A is a color rendition of FIG. 11.
FIG. 12 depicts a second enhanced image of FIG. 4 generated with an embodiment of the present invention to emphasize cars, parked and moving, and colored rooftops having some commonality with automotive paint.
FIG. 12A is a color rendition of FIG. 12.
FIG. 13 depicts a third enhanced image of FIG. 4 generated with an embodiment of the present invention to emphasize roads and parking lots.
FIG. 13A is a color rendition (no gray scale) of FIG. 13.
FIG. 14 depicts a fourth enhanced image of FIG. 4 generated with an embodiment of the present invention to emphasize painted surfaces of the same approximate color.
FIG. 14A is a color rendition of FIG. 14.
FIG. 15 depicts a class map of FIG. 4, generated from data used to create FIGS. 11, 12, 13 and 14.
FIG. 15A is a color rendition of FIG. 15.
FIG. 16 is a flowchart representing a method of implementing data mining as used in select embodiments of the present invention.
DETAILED DESCRIPTIONIn select embodiments of the present invention, the expansion of a simple hyperspectral ratio with wavelet coefficients provides an easy, fast, and reliable method for analysis, categorization, or feature extraction (identification). In addition to the analysis of hyperspectral datasets, select embodiments of the present invention are sufficiently general for use with any high dimensional dataset.
Select embodiments of the present invention sample all band ratio combinations in hyperspectral data for rapid combinatorial computations that integrate wavelet and waveletvariogram techniques for better data anomaly filtering and detection. In select embodiments of the present invention, a very efficient software package display results in a form that is easily comprehensible for classification and feature extraction analysis. In select embodiments of the present invention, one is able to investigate if there are band ratios that provide potentially useful information in support of classification and feature extraction tasks.
In select embodiments of the present invention, all possible band ratio combinations need not be computed. Several factors contribute to yielding a small subset of ratios that provides viable information.
Refer to FIGS. 1 and 1A provided for illustration purposes only. FIG. 1 is a black and white version of the more readily discernible color FIG. 1A. Wavelength, nanometers (ηm), is plotted against radiance for most of the features in a hyperspectral data scene. Note that the reflectance values greater than 1799 ηm do not vary much, i.e., the spectral curve is flat with radiance values close to zero. Therefore, all ratios generated from bands between 1799 and 2500 ηm will provide essentially the same information and one ratio computation within any of these bands is sufficient.
As noted above, inherent disadvantages of a hyperspectral data set are the vast amount of data and contiguous bands highly correlated due to close proximity within the electromagnetic spectrum. For example, ratios involving adjacent bands generally provide the same information as ratios within those bands separated by two or three bands.
Refer to FIG. 2 showing two images 201, 202 generated using two different ratios from a hyperspectral data set. The image 201 on the reader's left was generated from the difference between band 109, (wavelength 1437.3 ηm) and band 110, (wavelength 1451.1 ηm) and the image 202 on the reader's right generated from the difference between band 109 and band 111 (wavelength 1464.9 ηm). The two images 201, 202 appear almost identical so both are likely to provide the same information. The similarity between these images 201, 202 suggests that computing and analyzing ratios involving adjacent bands and ratios involving bands separated by two bands yield the same information, so computation of ratios separated by two bands is sufficient in this case.
Another factor that reduces the number of ratios needed, is the number of features in a hyperspectral scene. Normally the number of features of interest is less than the total number of features in the scene. Thus, in many situations most features of interest are extracted employing the analysis of but a few band ratios.
Unlike the work to derive either VI or NDVI in which the user is interested mostly in the numerical value of a ratio, in select embodiments of the present invention, the image display of the resulting ratio data is desired. Thus, of interest is the numerator of the NDVI approaching zero (0) for a particular vegetation feature, such as trees. This occurs when the radiance of the spectra of this prespecified feature, e.g., trees, is the same for two different wavelengths, i.e., the difference in radiance equals zero, and the radiance of all other features in the dataset are different (nonzero) at these wavelengths. If these conditions are true and the resulting dataset is displayed as an image, the pixels for the trees will appear black and all other pixels a shade of gray. It is then easy to identify the trees in the image. The converse is also useful. If the radiance of the spectra for the trees is different in two wavelengths and the radiance of the spectra for all other features is the same in the same two wavelengths, then the image data will appear with the trees a shade of gray and all other features as black. The numerator does not have to vanish totally, but as long as there is separation in radiance between one feature and all the other features, then that one feature may be extracted easily.
A Generalized Difference Feature Index (GDFI), computed using Daubechies wavelets, automatically generates a large sequence of generalized band ratio images.
Ordinary band ratio methods for the ready identification of vegetation, cloud cover, and other features, usually take the difference of two bands divided by their sum. For example, for bands 5 and 7, in LANDSAT® multispectral imagery the NDVI for each pixel in the image is computed by:<maths id="MATHUS00002" num="2"><math overflow="scroll"><mtable><mtr><mtd><mrow><mi>NDVI</mi><mo>=</mo><mfrac><mrow><mo>(</mo><mrow><msub><mi>z</mi><mn>7</mn></msub><mo></mo><msub><mi>z</mi><mn>5</mn></msub></mrow><mo>)</mo></mrow><mrow><mo>(</mo><mrow><msub><mi>z</mi><mn>7</mn></msub><mo>+</mo><msub><mi>z</mi><mn>5</mn></msub></mrow><mo>)</mo></mrow></mfrac></mrow></mtd><mtd><mrow><mo>(</mo><mn>5</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths>where z<sub>i</sub>, 1≦i≦7, corresponds to the spectral signature of the image. By scaling each of the terms in the numerator and denominator by<maths id="MATHUS00003" num="3"><math overflow="scroll"><mrow><mfrac><mn>1</mn><msqrt><mn>2</mn></msqrt></mfrac><mo>,</mo></mrow></math></maths>the ratio of Haar's wavelet high frequency content over its low frequency content is computed. That is, for Eqn. (5):<maths id="MATHUS00004" num="4"><math overflow="scroll"><mtable><mtr><mtd><mrow><mi>NDVI</mi><mo>=</mo><mfrac><mrow><mo>(</mo><mrow><mfrac><msub><mi>z</mi><mn>7</mn></msub><msqrt><mn>2</mn></msqrt></mfrac><mo></mo><mfrac><msub><mi>z</mi><mn>5</mn></msub><msqrt><mn>2</mn></msqrt></mfrac></mrow><mo>)</mo></mrow><mrow><mo>(</mo><mrow><mfrac><msub><mi>z</mi><mn>7</mn></msub><msqrt><mn>2</mn></msqrt></mfrac><mo>+</mo><mfrac><msub><mi>z</mi><mn>5</mn></msub><msqrt><mn>2</mn></msqrt></mfrac></mrow><mo>)</mo></mrow></mfrac></mrow></mtd><mtd><mrow><mo>(</mo><mn>6</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths>
This can be considered a first order scaled derivative. The data produced by applying this ratio accentuates discontinuities and suppresses slowlyvarying data within a spectral signature.
At any pixel, Eqn. (6) may be generalized by applying the ratio throughout the components of the spectral signature, z, yielding a simplified relationship describing the Generalized Difference Feature Index (GDFI) of the present invention,<maths id="MATHUS00005" num="5"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><msub><mi>GDFI</mi><mn>2</mn></msub><mo></mo><mrow><mo>(</mo><mrow><mi>i</mi><mo>,</mo><mi>t</mi></mrow><mo>)</mo></mrow></mrow><mo>=</mo><mfrac><mrow><mrow><msub><mi>h</mi><mn>0</mn></msub><mo></mo><msub><mi>z</mi><mi>i</mi></msub></mrow><mo>+</mo><mrow><msub><mi>h</mi><mn>1</mn></msub><mo></mo><msub><mi>z</mi><mrow><mi>i</mi><mo>+</mo><mi>t</mi></mrow></msub></mrow></mrow><mrow><mrow><msub><mi>g</mi><mn>0</mn></msub><mo></mo><msub><mi>z</mi><mi>i</mi></msub></mrow><mo>+</mo><mrow><msub><mi>g</mi><mn>1</mn></msub><mo></mo><msub><mi>z</mi><mrow><mi>i</mi><mo>+</mo><mi>t</mi></mrow></msub></mrow></mrow></mfrac></mrow></mtd><mtd><mrow><mo>(</mo><mn>7</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths>where:
 i refers to band i of a data collector such as LANDSAT®,
 t is the lag between bands, i.e., number of bands skipped
 h<sub>0</sub>, h<sub>1 </sub>are the high frequency coefficients
 g<sub>0</sub>, g<sub>1 </sub>are the low frequency coefficients, and in particular
 h<sub>0</sub>=1/√{square root over (2)}, h<sub>1</sub>=−1/√{square root over (2)}, g<sub>0</sub>=1/√{square root over (2)}, and g<sub>1</sub>=1/√{square root over (2)}
The subscript 2 in GDFI<sub>2 </sub>refers to the number of filter coefficients used in the ratio. For example, using LANDSAT® imagery yields the NDVI by:<FORM>NDVI=−GDFI<sub>2</sub>(5,2) (8)</FORM>where for LANDSAT® imagery conventionally used to yield the NVDI, i is 5 (Band 5) and t, lag, is 2 such that the difference between Bands 7 and 5 is divided by the sum of Bands 5 and 7 to yield the NDVI.
The method may be further generalized to include other wavelet coefficients with a higher number of vanishing moments. A natural extension of Eqn. (7) accounts for more filter coefficients, such as Daubechies high and low frequency filter coefficients. Eqn. (7) can be written as:<maths id="MATHUS00006" num="6"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><msub><mi>GDFI</mi><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow></msub><mo></mo><mrow><mo>(</mo><mrow><mi>i</mi><mo>,</mo><mi>t</mi></mrow><mo>)</mo></mrow></mrow><mo>=</mo><mfrac><mrow><mrow><msub><mi>h</mi><mi>o</mi></msub><mo></mo><msub><mi>z</mi><mi>i</mi></msub></mrow><mo>+</mo><mrow><msub><mi>h</mi><mn>1</mn></msub><mo></mo><msub><mi>z</mi><mrow><mi>i</mi><mo>+</mo><mi>t</mi></mrow></msub></mrow><mo>+</mo><mi>…</mi><mo></mo><mstyle><mtext> </mtext></mstyle><mo>+</mo><mrow><msub><mi>h</mi><mrow><mo>(</mo><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow><mo>)</mo></mrow></msub><mo></mo><msub><mi>z</mi><mrow><mi>i</mi><mo>+</mo><mrow><mrow><mo>(</mo><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow><mo>)</mo></mrow><mo></mo><mi>t</mi></mrow></mrow></msub></mrow></mrow><mrow><mrow><msub><mi>g</mi><mi>o</mi></msub><mo></mo><msub><mi>z</mi><mi>i</mi></msub></mrow><mo>+</mo><mrow><msub><mi>g</mi><mn>1</mn></msub><mo></mo><msub><mi>z</mi><mrow><mi>i</mi><mo>+</mo><mi>t</mi></mrow></msub></mrow><mo>+</mo><mi>…</mi><mo></mo><mstyle><mtext> </mtext></mstyle><mo>+</mo><mrow><msub><mi>g</mi><mrow><mo>(</mo><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow><mo>)</mo></mrow></msub><mo></mo><msub><mi>z</mi><mrow><mi>i</mi><mo>+</mo><mrow><mrow><mo>(</mo><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow><mo>)</mo></mrow><mo></mo><mi>t</mi></mrow></mrow></msub></mrow></mrow></mfrac></mrow></mtd><mtd><mrow><mo>(</mo><mn>9</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths>where n corresponds to the number of vanishing moments of the wavelet. Thus, In select embodiments of the present invention, the ratio method is employed for feature extraction (identification) and classification and a procedure for data mining is used therewith as described below in a flowchart that represents a process used in a select embodiment of the present invention.
Daubechies maxflat filters comprise an even number of coefficients where the high frequency filters satisfy n vanishing moments according to:<FORM>h<sub>o</sub>*0<sup>k</sup>+h<sub>1</sub>*1<sup>k</sup>+ . . . +h<sub>2n−1</sub>*(2n−1)<sup>k</sup>=0 (10)</FORM>and the low frequency coefficients, g, satisfy:<FORM>g<sub>o</sub>+g<sub>1</sub>+ . . . +g<sub>2n−1</sub>=√{square root over (2)} (11)</FORM>for 0≦k≦n−1.
Eqns. (10) and (11) establish the criteria that a wavelet's high frequency, h<sub>k</sub>, and low frequency, g<sub>k</sub>, coefficients must satisfy to form a wavelet. For example, Eqn. (10) for a wavelet with three (3) vanishing moments, i.e., n=3, would be written as three separate expressions:<FORM>h<sub>0</sub>*0<sup>0</sup>+h<sub>1</sub>*1<sup>0</sup>+h<sub>2</sub>*2<sup>0</sup>+h<sub>3</sub>*3<sup>0</sup>+h<sub>4</sub>*4<sup>0</sup>+h<sub>5</sub>*5<sup>0</sup>=0 (zero moment, or k=0) (10a)</FORM><FORM>h<sub>0</sub>*0<sup>1</sup>+h<sub>1</sub>*1<sup>1</sup>+h<sub>2</sub>*2<sup>1</sup>+h<sub>3</sub>*3<sup>1</sup>+h<sub>4</sub>*4<sup>1</sup>+h<sub>5</sub>*5<sup>1</sup>=0 (first moment, or k=0) (10b)</FORM><FORM>h<sub>0</sub>*0<sup>2</sup>+h<sub>1</sub>*1<sup>2</sup>+h<sub>2</sub>*2<sup>2</sup>+h<sub>3</sub>*3<sup>2</sup>+h<sub>4</sub>*4<sup>2</sup>+h<sub>5</sub>*5<sup>2</sup>=0 (second moment, or k=2) (10c)</FORM>
0<sup>0 </sup>is undefined, but for this case it is equal to 1.
These coefficients, g<sub>k </sub>and h<sub>k</sub>, are used to compute the GDFI (generalized ratios) of derived bands (indices) as defined in Eqn. (9). The sequence of GDFI “bands” is then used to identify features of interest in a scene, for example. Individual GDFI bands are linked to specific features of interest and, once linked, are used to quickly and efficiently process data to “identify” or “highlight” only those features of interest.
A desired image with specific features “highlighted” is provided by employing a specific GDFI band(s) ratio and multiplying it by an appropriate constant, such as the Daubechies coefficients. In select embodiments of the present invention, a feature or features becomes very bright, e.g., appearing as a distinct color or shade lighter than the remainder of the image. Of course, an increase in the number of band ratios in the image increases the complexity of the image with a concomitant need for higher coefficients, i.e., higher order wavelets. Thus, there is a limit at which so many of the pixels are highlighted as to make the image no more valuable than the original unmodified image. Thus, it is appropriate to employ the least ordered wavelet that enables a select few features to be readily identified.
Vanishing moments are related to scaled derivatives. Mallat, S., A Wavelet Tour of Signal Processing, Academic Press, San Diego, Calif. 1998. Due to the wavelets' higher number of vanishing moments, as with Eqn. (5), this technique will suppress slowly varying regions of the image or higherorder polynomials in the spectra while also accentuating discontinuities. Bosch et al. note that for a fixed value of the lag, t, the sum of the squares of the numerator of Eqn. (9) scaled by<maths id="MATHUS00007" num="7"><math overflow="scroll"><mfrac><mn>1</mn><mrow><mo>(</mo><mrow><mi>m</mi><mo></mo><mrow><mrow><mo>(</mo><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow><mo>)</mo></mrow><mo></mo><mi>t</mi></mrow></mrow><mo>)</mo></mrow></mfrac></math></maths>for all values of i, corresponds to the datamining waveletvariogram function ŷ(t) such that:<maths id="MATHUS00008" num="8"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><mover><mi>γ</mi><mo>^</mo></mover><mo></mo><mrow><mo>(</mo><mi>t</mi><mo>)</mo></mrow></mrow><mo>=</mo><mrow><mfrac><mn>1</mn><mrow><mo>(</mo><mrow><mi>m</mi><mo></mo><mrow><mrow><mo>(</mo><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow><mo>)</mo></mrow><mo></mo><mi>t</mi></mrow></mrow><mo>)</mo></mrow></mfrac><mo></mo><mrow><munderover><mo>∑</mo><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mrow><mi>m</mi><mo></mo><mrow><mrow><mo>(</mo><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow><mo>)</mo></mrow><mo></mo><mi>t</mi></mrow></mrow></munderover><mo></mo><mstyle><mtext> </mtext></mstyle><mo></mo><msup><mrow><mo>[</mo><mrow><mrow><msub><mi>h</mi><mi>o</mi></msub><mo></mo><msub><mi>z</mi><mi>i</mi></msub></mrow><mo>+</mo><mrow><msub><mi>h</mi><mn>1</mn></msub><mo></mo><msub><mi>z</mi><mrow><mi>i</mi><mo>+</mo><mi>t</mi></mrow></msub></mrow><mo>+</mo><mrow><mi>…</mi><mo></mo><mstyle><mtext> </mtext></mstyle><mo></mo><msub><mi>h</mi><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow></msub><mo></mo><msub><mi>z</mi><mrow><mi>i</mi><mo>+</mo><mrow><mrow><mo>(</mo><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow><mo>)</mo></mrow><mo></mo><mi>t</mi></mrow></mrow></msub></mrow></mrow><mo>]</mo></mrow><mn>2</mn></msup></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>12</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths>where:
 m is the number of spectral bands in the data set,
 n is the number of vanishing moments in the filter wavelet, and
 t is the lag.Bosch, E. H., M. A. Oliver and R. Webster, Wavelets and the Generalization of the Variogram, Mathematical Geology, Volume 36, No. 2, pages 147186, February 2004.
The experimental variogram, {circumflex over (γ)}(t), as used in geostatistics, is obtained when n=1 (Haar wavelet filter). The experimental variogram, {circumflex over (γ)}(t), analyzes spatial properties of data at different scales.
The GDFI<sub>2n </sub>is applied as a data mining technique to expose (identify) features of interest. The resulting generalization produces band ratios that depend on the number of spectral bands, the lag parameter, and the length of the Daubechies maxflat filter. Through employing the band ratio images as a datamining tool one is able to rapidly display (identify) those prespecified features enhanced by an embodiment of the present invention. The desired prespecified features may be viewed as grayscale images but, as with images displayed using the NDVI, color composites may also be formed using an embodiment of the present invention. Hyperspectral imagery provides more flexibility, given the availability of a higher number of spectral bands. This is why embodiments of the present invention apply the filtered ratios first and then display the images.
As with the experimental variogram, {circumflex over (γ)}(t), this technique depends on the assumption that data points are regularly spaced, a generally valid assumption since spectral bands from the employed sensors approximate regular spacing.
EXAMPLE IData were collected with a Hyperspectral Digital Collection Experiment (HYDICE) airborne sensor. This sensor uses a 320×210 array with multiple gain regions to support imaging at 4002500 ηm. The sensor is flown at 5000 to 25000 ft and achieves spatial resolution of about 0.8 to 4 m. Basedow, R. W., D. C. Carmer, and M. E. Anderson, Hydice: Operational Status, Proceedings of the SPIE, Vol. 2480, May 1994.
Data were collected over Copperas Cove, Tex. and the neighboring vicinity. The aircraft flew at an altitude of 20,000 ft to provide a spatial resolution of three (3) meters in 210 spectral bands from 398.52499.2 ηm. Twelve (12) features were identified in the hyperspectral scene: buildings 301, roads 302, soil 303, rooftops 304, canvas (not shown separately), rubber (not shown separately), metal (not shown separately), shrubs 305, plastic, grass 306, trees 307, and calibration panels (not shown separately). Refer to FIG. 3 depicting the hyperspectral composite class map of the Copperas Cove area in gray scale and to FIG. 3A showing the same composite class map in color.
EXAMPLE IIData were also collected over Yuma City, Ariz. The aircraft flew at an altitude of 5000 ft to provide a spatial resolution of one (1) meter in the same number of bands as in Example 1. However, for this example only 204 bands were employed to image. These data are not calibrated. Eight (8) features were identified in the scene including: buildings 401, roads 402, soil 403, rooftops 404, shrubs 405, grass 406, trees 407, and swimming pools 408. FIG. 4 is a grayscale composite of the imaged area while FIG. 4A is the same Yuma City composite in color.
In the data sets for Examples I and II the interval between bands is not constant. Table I below summarizes the first three (3) and last five (5) wavelengths measured for each data set. The plot of the wavelength interval for Copperas Cove is a monotonic curve starting with an approximate interval of 3.3 and increasing to about 8.3 ηm. The same plot for Yuma City data is a step function with the interval a constant amount for several wavelengths followed by a jump. The two data sets are different although both were collected with the same sensor. The adjacent bands of each data set, even with the jump in the Yuma City data, are reasonably close enough to preclude any discontinuity problems.<tables id="TABLEUS00001" num="1"><table frame="none" colsep="0" rowsep="0"><tgroup align="left" colsep="0" rowsep="0" cols="1"><colspec colname="1" colwidth="217PT" align="center"/><thead><row><entry namest="1" nameend="1" align="center">TABLE 1</entry></row></thead><tbody valign="top"><row><entry/></row><row><entry namest="1" nameend="1" align="center" rowsep="1"/></row><row><entry>Band and Interval Relationship for Two Sites.</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="1" colwidth="112PT" align="center"/><colspec colname="2" colwidth="105PT" align="center"/><tbody valign="top"><row><entry>COPPERAS COVE</entry><entry>YUMA CITY</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="6"><colspec colname="1" colwidth="42PT" align="center"/><colspec colname="2" colwidth="42PT" align="center"/><colspec colname="3" colwidth="28PT" align="center"/><colspec colname="4" colwidth="35PT" align="center"/><colspec colname="5" colwidth="42PT" align="center"/><colspec colname="6" colwidth="28PT" align="center"/><tbody valign="top"><row><entry/><entry>Wavelength</entry><entry>Interval</entry><entry/><entry>Wavelength</entry><entry>Interval</entry></row><row><entry>Band No.</entry><entry>(ηm)</entry><entry>(ηm)</entry><entry>Band No.</entry><entry>(ηm)</entry><entry>(ηm)</entry></row><row><entry namest="1" nameend="6" align="center" rowsep="1"/></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="6"><colspec colname="1" colwidth="42PT" align="char" char="."/><colspec colname="2" colwidth="42PT" align="char" char="."/><colspec colname="3" colwidth="28PT" align="center"/><colspec colname="4" colwidth="35PT" align="char" char="."/><colspec colname="5" colwidth="42PT" align="char" char="."/><colspec colname="6" colwidth="28PT" align="char" char="."/><tbody valign="top"><row><entry>1</entry><entry>398.544</entry><entry>3.296</entry><entry>1</entry><entry>460.000</entry><entry>6.0</entry></row><row><entry>2</entry><entry>401.840</entry><entry>3.306</entry><entry>2</entry><entry>464.000</entry><entry>6.0</entry></row><row><entry>3</entry><entry>405.146</entry><entry>3.334</entry><entry>3</entry><entry>468.000</entry><entry>6.0</entry></row><row><entry>206</entry><entry>2466.270</entry><entry>8.270</entry><entry>200</entry><entry>2230.000</entry><entry>19.0</entry></row><row><entry>207</entry><entry>2474.540</entry><entry>8.250</entry><entry>201</entry><entry>2249.000</entry><entry>9.0</entry></row><row><entry>208</entry><entry>2482.790</entry><entry>8.209</entry><entry>202</entry><entry>2258.000</entry><entry>9.0</entry></row><row><entry>209</entry><entry>2491.000</entry><entry>8.169</entry><entry>203</entry><entry>2267.000</entry><entry>9.0</entry></row><row><entry>210</entry><entry>2499.169</entry><entry>—</entry><entry>204</entry><entry>2276.000</entry><entry>—</entry></row><row><entry namest="1" nameend="6" align="center" rowsep="1"/></row></tbody></tgroup></table></tables>
Several prespecified desired features were extracted (identified) from scenes by using an embodiment of the present invention. Daubechies filtering wavelets with two (2), four (4) and eight (8) filter coefficients were used to generate numerous “differencesum ratios.” Thesholding was applied to separate the prespecified features from the rest of the image and produce a binary image. Normally, thresholding is not amenable to automated computation. Thus, the threshold value is manually adjusted until a satisfactory result is obtained. Alternatively, thresholding is done by employing some statistical measure, such as a standard deviation from a class mean. Both of these thresholding methods can be timeconsuming.
Refer to FIG. 5, an “enhanced” grayscale image of FIG. 3 and its color rendition FIG. 5A, generated from Eqn. (9) where n=1, i=17, and t=6. These values equate to the differencesum ratio of bands 17 and 23 employing Daubechies wavelets and two (2) filter coefficients. FIG. 5 has been remapped with histogram equalization for better visual display. The light gray shading (green in FIG. 5A) represents vegetation consisting of trees, shrubs, and bushes, but no grass. This is consistent with ground truth information and with the same scene as represented in FIGS. 3 and 3A. The very dark areas represent trees, shrubs, bushes or a combination thereof.
Refer to FIG. 6, an “enhanced” gray scale image of FIG. 3 and its color rendition FIG. 6A, generated from Eqn. (9), where n=1, i=120, t=9. These values equate to the differencesum ratio of bands 120 and 129 employing Daubechies wavelets and two (2) filter coefficients. The light shaded areas (blue in a color rendition) represent buildings. These buildings correlate very well with the buildings identified in FIG. 3 and with ground truth. Buildings are not a detectable class in hyperspectral classification, but in this case all the light colored pixels have some commonality. One possibility is a delicate combination of asphalt rooftops with just enough paint to exclude asphalt parking lots and roads.
Refer to FIG. 7, an “enhanced” gray scale image of FIG. 3 and its color rendition FIG. 7A, generated from Eqn. (9) where n=1, i=153, t=6. These values equate to the differencesum ratio of bands 153 and 159 employing Daubechies wavelets and two (2) filter coefficients. The light shading (light yellowish brown in FIG. 7A) represents the grass shown in FIG. 3. At three (3) meterresolution most of the pixels are not homogeneous but may be a mix of grass, leaves, topsoil, and some tree branches in different combinations.
Refer to FIG. 8, an “enhanced” gray scale image of FIG. 3 and its color rendition FIG. 8A, a ground truth photo of a typical grassy area in Copperas Cove. These grassy areas, as depicted in FIGS. 8 and 8A, are a mix of low growing vegetation, grass, branches, and soil.
Refer to FIG. 9, an “enhanced” grayscale image of FIG. 3 and its color rendition FIG. 9A, generated from Eqn. (9) where n=1, i=18, t=5. These values equate to the differencesum ratio of bands 18 and 23 employing Daubechies wavelets and two (2) filter coefficients. The light shading (red in FIG. 9A) represents dirt roads, concrete, and bright rooftops of the scene from FIG. 3. The complex constitution of some roads, e.g., debris, tire marks, and the like may cause some roads to appear to have disconnected segments.
Refer to FIG. 10, an “enhanced” grayscale image of FIG. 3 and its color rendition FIG. 10A, a classification (“class”) map of Copperas Cove, generated from data used to create FIGS. 5, 6, 7, and 9 and their respective color renditions, FIGS. 5A, 6A, 7A, and 9A. No attempt was made to optimize color, but this can be done. In the color rendition of FIG. 10A, the major features are well separated, viz., trees, shrubs and bushes 1004 (FIG. 10) are green, dirt roads 1005 (FIG. 10) are orange, grass and soil 1001 (FIG. 10) are brownish, and asphalt 1002 (FIG. 10) is black. Purple pixels represent bright rooftops 1003 (FIG. 10). There are some pixels that did not get classified, but most of them are sparsely scattered throughout the image and most are probably mixed pixels such as those that may represent mixed vegetation, i.e., a mix of topsoil, grass, trees, shrubs, and bushes. There may be other features that could be extracted (identified), e.g., swimming pools, but they are too small to be clearly identified in the image of FIGS. 4 and 4A.
Ground truth at Copperas Cove provided five (5) other features: rubber, plastic, metal, canvas, and calibration panels. These were not clearly identified with an embodiment of the present invention. All these features, however, are very small in size and in number and are very challenging to detect under any circumstances. Calibration panels are detectable, but they are so small that it is not possible to conclusively identify them and normally these pixels are dismissed as artifacts.
Refer to FIG. 11, an “enhanced” grayscale image of FIG. 4, and its color rendition FIG. 11A, generated from Eqn. (9), where n=2, i=26, and t=16. The relevant bands are 26, 42, 58, and 74 with Daubechies wavelets, four (4) filter coefficients and thresholding. The light shaded portions of FIG. 11 and the corresponding green of FIG. 11A, represent trees, shrubs, bushes, and grass. The trees, shrubs, bushes, and grass in the image correlate very well with FIG. 4 and with ground truth, given that FIGS. 11 and 11A are not exact duplicate representations of the area of FIG. 4, i.e., they correlate to only a part of the area of FIG. 4.
Refer to FIG. 12, an “enhanced” grayscale image of FIG. 4, and its color rendition FIG. 12A, generated from Eqn (9) where n=2, i=135, and t=16. The relevant bands are 135, 151, 167, and 183 with Daubechies wavelets, four (4) filter coefficients, and thresholding. The small lightshaded signatures of FIG. 12 (blue signatures in the color rendition of FIG. 12A) represent parked and moving vehicles, and the larger lightshaded signatures of FIG. 12 (blue signatures in the color rendition of FIG. 12A) are rooftops that have some commonality with automotive paint. The large solid signature 1201 (FIG. 12) is a commercial building having an asphalt rooftop with some metallic paint on it. Ground truth verified that the other lightshaded signatures (blue signatures in the color rendition of FIG. 12A) have painted surfaces. Not all the vehicles in the scene have been extracted (identified) because different color shades have substantially different spectra. Vehicles and rooftops are not considered to constitute a unique, independent class in hyperspectral classification unless all have identical surface paint. The spectral response for wood and metal painted the same color is very nearly the same. Evans, T., P. Henley, and E. Gonzalez, Desert Radiance II: Supplemental Report, submitted to Hyperspectral MASINT Support to Military Operations (HYMSMO) Program Office, October 1995.
Refer to FIG. 13, an “enhanced” grayscale image of FIG. 4, and its color (no gray scale) rendition FIG. 13A, generated from Eqn. (9) where n=1, i=15, t=16. These values equate to the differencesum ratio of bands 15 and 31 employing Daubechies wavelets and two (2) filter coefficients. The white areas represent roads and parking lots of FIG. 4. The composition of roads and parking lots in this area is gravel, concrete, and asphalt in different combinations. All these materials have very similar spectral properties so that in most cases, roads, whether all one material or a combination of several, will appear the same. Evans et al. (1995). The white areas correlate well with the roads, parking lots, and some gravel covered or asphalt, not bright, rooftops of FIG. 4.
Refer to FIG. 14, an “enhanced” grayscale image of FIG. 4, and its color rendition FIG. 14A, generated from Eqn. (9) where n=1, i=32, t=6. These values equate to the differencesum ratio of bands 32 and 38 employing Daubechies wavelets and two (2) filter coefficients. The light shaded areas (pink in the color rendition of FIG. 14A) represent surfaces painted the same color, e.g., roofing edges. Most of these surfaces are on rooftops and the small dots are vehicles. The horizontal broken line in both FIGS. 14 and 14A is an artifact.
Refer to FIG. 15, an “enhanced” grayscale image and its color rendition FIG. 15A, is a rudimentary class map of the Yuma City image of FIG. 4, generated from FIGS. 11, 12, 13, and 14 and 11A, 12A, 13A, and 14A, respectively. Vegetation 1501, including grass, is shown in green in the color rendition of FIG. 15A, roads, parking lots, and some rooftops 1502 are gray in FIG. 15A, colored rooftops 1503 are blue in FIG. 15A, colored sides of buildings 1504 are orange in FIG. 15A, and everything that was not classified is black in FIG. 15A.
A single prespecified feature may be extracted (identified) with different combinations of bands. Table II below shows some features and the bands that were used for the extraction (identification) and associated wavelengths. The fourth column shows the number of coefficients used with the Daubechlies wavelet employed with this particular embodiment of the present invention. For example, D4 stands for Daubechies wavelets with four (4) filter coefficients. Lag is the interval between the bands, i.e., number of skipped bands, used in this embodiment of the present invention. The sequence of differencesum ratio images generated with the same lag and same filter wavelet constitutes an image cube. Often all prespecified features of interest may be extracted (identified) within one of these image cubes.<tables id="TABLEUS00002" num="2"><table frame="none" colsep="0" rowsep="0" pgwide="1"><tgroup align="left" colsep="0" rowsep="0" cols="1"><colspec colname="1" colwidth="273PT" align="center"/><thead><row><entry namest="1" nameend="1" align="center">TABLE 2</entry></row></thead><tbody valign="top"><row><entry/></row><row><entry namest="1" nameend="1" align="center" rowsep="1"/></row><row><entry>Extracted (identified) Features</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="6"><colspec colname="1" colwidth="49PT" align="left"/><colspec colname="2" colwidth="35PT" align="center"/><colspec colname="3" colwidth="49PT" align="center"/><colspec colname="4" colwidth="28PT" align="center"/><colspec colname="5" colwidth="21PT" align="center"/><colspec colname="6" colwidth="91PT" align="left"/><tbody valign="top"><row><entry/><entry/><entry>Wavelengths</entry><entry/><entry/><entry/></row><row><entry>Scene</entry><entry>Bands</entry><entry>(ηm)</entry><entry>Wavelet</entry><entry>Lag</entry><entry>Feature</entry></row><row><entry namest="1" nameend="6" align="center" rowsep="1"/></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="6"><colspec colname="1" colwidth="49PT" align="left"/><colspec colname="2" colwidth="35PT" align="center"/><colspec colname="3" colwidth="49PT" align="center"/><colspec colname="4" colwidth="28PT" align="center"/><colspec colname="5" colwidth="21PT" align="char" char="."/><colspec colname="6" colwidth="91PT" align="left"/><tbody valign="top"><row><entry>Copperas Cove</entry><entry> 1/17</entry><entry>398.5/455.7</entry><entry>D4</entry><entry>16</entry><entry>Vegetation very dark</entry></row><row><entry>Copperas Cove</entry><entry> 7/23</entry><entry>418.6/481.0</entry><entry>D4</entry><entry>16</entry><entry>Swimming Pools very dark</entry></row><row><entry>Copperas Cove</entry><entry>12/15</entry><entry>436.5/447.8</entry><entry>D2</entry><entry>3</entry><entry>Roads, parking lots,</entry></row><row><entry/><entry/><entry/><entry/><entry/><entry>rooftops bright</entry></row><row><entry>Copperas Cove</entry><entry>52/58</entry><entry>668.3/727.7</entry><entry>D2</entry><entry>6</entry><entry>Vegetation very bright</entry></row><row><entry>Yuma City</entry><entry>81/88</entry><entry>1030.0/1134.0</entry><entry>D4</entry><entry>7</entry><entry>Roads, rooftops very bright</entry></row><row><entry>Yuma City</entry><entry>90/97</entry><entry>1164.0/1267.0</entry><entry>D4</entry><entry>7</entry><entry>Vegetation very bright</entry></row><row><entry>Yuma City</entry><entry>42/58</entry><entry>588.0/729.0</entry><entry>D2</entry><entry>16</entry><entry>Swimming pools very bright</entry></row><row><entry>Yuma City</entry><entry>112/128</entry><entry>1480.0/1687.0</entry><entry>D2</entry><entry>16</entry><entry>Vegetation very bright</entry></row><row><entry namest="1" nameend="6" align="center" rowsep="1"/></row></tbody></tgroup></table></tables>
EXAMPLE IIIRefer to FIG. 16, a sample flow diagram for a process used in select embodiments of the present invention to perform data mining to streamline the processing of hyperspectral data in those cases in which specific features are to be identified or highlighted. To start the process at step 11601 the prespecified limit of the wavelet array index, K, is selected and set and start values of wavelet array index, K, and lag, T, are set to zero and 1, respectively. Next at step 21602, the limit for lag is set. Having set these two limits, step 31603 involves reading in the hyperspectral data set. Step 41604 discards those bands that contain compromised (bad) data. (Although this step is taken early in the process to reduce the amount of data handled early on, “bad bands” would be eliminated at step 101610 below in any case.) The wavelet array index, K, is then incremented appropriately in step 51605. In step 61606, the lag, T, is incremented by 1 and step 71607 calculates the ratios that are appropriate to generate indices in step 81608. A “cube” of the indices is generated in step 91609 to provide the source from which indices for extracting specific features are selected in step 101610. It is in step 101610 that the number of bands are reduced. It is possible to have started with hundreds of bands and after step 101610, many derived bands are eliminated. One advantage of this process is that there is no need to determine if a band is “good” or “bad,” just appropriate to identify the specified structure. In step 111611, a threshold is applied and only those indices meeting the threshold are selected for further use. The thresholded indices are saved in step 121612 and two limits are applied to therm. In step 131613, a lag limit is applied. The lag limit, t, is defined as<maths id="MATHUS00009" num="9"><math overflow="scroll"><mtable><mtr><mtd><mrow><mn>1</mn><mo>≤</mo><mi>t</mi><mo>≤</mo><mrow><mo>(</mo><mrow><mi>integer</mi><mo>(</mo><mfrac><mrow><mi>m</mi><mo></mo><mn>1</mn></mrow><mrow><mrow><mn>2</mn><mo></mo><mi>n</mi></mrow><mo></mo><mn>1</mn></mrow></mfrac><mo>)</mo></mrow><mo>)</mo></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>13</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths>Where:
 m=number of bands in the dataset, and
 n=number of vanishing moments.
If an index meets the lag limit, it is passed to step 141614 where a wavelet order limit is applied. Upon passing both the wavelet order and lag limits, the index is “approved” for the desired feature extraction and the process is ended 1616. If the lag limit is not met in step 131613, the process is recycled to step 61605, the lag is incremented by one and the process reiterated from step 71607 until once again reaching step 131613 whereupon a determination of whether it meets the lag limit is made. This process is repeated until the lag limit is met at step 131613 and then passed to step 141614 to determine if the wavelet order limit is met. If the wavelet order limit is not met, the process goes to step 151615 where the lag is reset to zero and the process iterated from step 51605 by incrementing the wavelet order by one and continuing the process to step 141614 until the wavelet order limit is met at which time the process is ended 1616. This process reduces the number of bands to be considered, making the feature extraction process much more efficient than previously possible using more complex methods.
In sum, the band ratio method of select embodiments of the present invention reduces computation cost in classification and feature extraction tasks. The shear number of possible ratios is, as stated, enormous. Selective band utilization reduces them to a manageable number, although the introduction of wavelets may increase computations several fold. There are several more wavelets that may be employed with select embodiments of the present invention, such as Vaidyanathan, Coiflet, Beylkin, and Symmlet, among others.
The abstract of the disclosure is provided to comply with the rules requiring an abstract that will allow a searcher to quickly ascertain the subject matter of the technical disclosure of any patent issued from this disclosure. 37 CFR § 1.72(b). Any advantages and benefits described may not apply to all embodiments of the invention.
While the invention has been described in terms of some of its embodiments, those skilled in the art will recognize that the invention can be practiced with modifications within the spirit and scope of the appended claims. For example, although specific types of image processing are discussed, other similar configurations or methods, including those that may have only some of the constituents or steps used in the above examples, may be suitable for extracting (identifying) prespecified features and thus fall within the ambit of a preferred embodiment of the present invention as provided in the claims herein.
In the claims, meansplusfunction clauses are intended to cover the structures described herein as performing the recited function and not only structural equivalents, but also equivalent structures. Thus, although a nail and a screw may not be structural equivalents in that a nail employs a cylindrical surface to secure wooden parts together, whereas a screw employs a helical surface, in the environment of fastening wooden parts, a nail and a screw may be equivalent structures.
Thus, it is intended that all matter contained in the foregoing description or shown in the accompanying drawings shall be interpreted as illustrative rather than limiting, and the invention should be defined only in accordance with the following claims and their equivalents.