System and method for hyper-spectral analysis

• US 20050254709A1
• Filed: 03/07/2005
• Published: 11/17/2005
• Est. Priority Date: 04/09/1999
• Status: Active Grant

• Current Assignee: Plain Sight Systems Incorporated (Water Intelligence Plc)
• Original Assignee: Plain Sight Systems Incorporated (Water Intelligence Plc)
• Inventors: Coppi, Andreas, Fateley, William G., Geshwind, Frank, Deverse, Richard A., Coifman, Ronald R., Davis, Gustave L., Warner, Frederick J.
• Granted Patent: US 7,219,086 B2
• Time in Patent Office: Days
• US Class Current: 382/182
• CPC Class Codes: G01J 3/02 Details
• CPC Class Codes: G01J 3/0208 using focussing or collimating elements, e.g. lenses or mirrors; performing aberration correction
• CPC Class Codes: G01J 3/021 using plane or convex mirrors, parallel phase plates, or particular reflectors
• CPC Class Codes: G01J 3/0218 using optical fibers
• CPC Class Codes: G01J 3/10 Arrangements of light sources specially adapted for spectrometry or colorimetry
• CPC Class Codes: G01J 3/28 Investigating the spectrum using colour filters G01J3/51
• CPC Class Codes: G01J 3/2823 Imaging spectrometer
• CPC Class Codes: G01J 3/36 Investigating two or more bands of a spectrum by separate detectors
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: G06V 20/13 Satellite images

Abstract

A hyper-spectral analysis method for characterizing or distinguishing diverse elements within hyper-spectral images. A plurality of patches of pixels from within the hyper-spectral images are extracted as being patches around pixels of the elements to be characterized or distinguished. The statistics of spectra for each patch of pixels are computed. A first classifier is computed from frequency-wise standard deviation of the spectra in each patch and a set of second classifiers are computed from principal components of the spectral in each patch. A combined classifier is computed based on the output of the first classifier and at least one of the second classifiers. The elements are characterized or distinguished based on the output of at least one of the classifiers, preferably the combined classifier.

Claims

1. A method of characterizing diverse elements within hyper-spectral images, comprising the steps of:

extracting a plurality of patches of pixels from within said hyper-spectral images as being patches around pixels of said elements to be characterized;

computing the statistics of selected spectral features for each patch of pixels;

computing a first classifier from feature-wise standard deviation of said selected spectral features in said each patch;

computing a set of second classifiers from principal components of said selected spectral features in said each patch;

computing a combined classifier based on the output of said first classifier and at least one of said second classifiers; and

characterizing said elements based on the output of at least one of said classifiers.

2. The method of claim 1, wherein said selected spectral features are spectrum values of each pixel in said each patch.

3. The method of claim 1, wherein the step of characterizing comprises the step of characterizing said elements based on the output of said combined classifier.

4. The method of claim 1, wherein said hyper-spectral images are hyper-spectral images of a tissue;

and wherein said step of characterizing comprises the step of characterizing tissue elements.

5. The method of claim 4, wherein the step of characterizing comprises the step of analyzing local variability of hyper-spectral tissue signatures for said each patch to characterize said tissue elements.

6. The method of claim 5, wherein said hyper-spectral tissue signatures comprises biological and non-biological variability;

and wherein the step of analyzing local variability comprises the step of removing said non-biological variability.

7. The method of claim 4, wherein the step of computing a first classifier comprises the step of determining features which discriminate between frequency-wise standard deviations of groups of normal nuclei and frequency-wise standard deviations of groups of abnormal nuclei.

8. The method of claim 4, wherein the step of computing a second set of classifiers comprises the step of determining features which discriminate between said principal components of normal groups and said principal components of abnormal groups.

9. A computer readable medium comprising code for characterizing diverse elements with hyper-spectral images, said code comprising instructions for:

extracting a plurality of patches of pixels from within said hyper-spectral images as being patches around pixels of said elements to be characterized;

computing the statistics of selected spectral features for each patch of pixels;

computing a first classifier from feature-wise standard deviation of said selected spectral features in said each patch;

computing a set of second classifiers from principal components of said selected spectral features in said each patch;

computing a combined classifier based on the output of said first classifier and at least one of said second classifiers; and

characterizing said elements based on the output of at least one of said classifiers.

10. System for characterizing diverse elements within hyper-spectral images, comprising:

an extracting module for extracting a plurality of patches of pixels from within said hyper-spectral images as being patches around pixels of said elements to be characterized;

a computing module for computing the statistics of selected spectral features for each patch of pixels, a first classifier from feature-wise standard deviation of said selected spectral features in said each patch, a set of second classifiers from principal components of said selected spectral features in said each patch, and a combined classifier based on the output of said first classifier and at least one of said second classifiers; and

a characterization module for characterizing said elements based on the output of at least one of said classifiers.