Self-designing intelligent signal processing system capable of evolutional learning for classification/recognition of one and multidimensional signals

• US 7,016,885 B1
• Filed: 08/28/2002
• Issued: 03/21/2006
• Est. Priority Date: 08/28/2001
• Status: Expired due to Fees

• Current Assignee: State University System of Florida
• Original Assignee: University of Central Florida Research Foundation Inc. (State University System of Florida)
• Inventors: Krishnan, Venkatesh, Mikhael, Wasfy B., Abdelwahab, Manal M.
• Primary Examiner(s): Knight, Anthony
• Assistant Examiner(s): Datskovskiy, Sergey
• Application Number:
  US10/230,425
• Time in Patent Office: 1,301 Days
• Field of Search: 706/20, 382/115, 382/118, 382/190, 382/156, 348/135
• US Class Current: 706/20
• CPC Class Codes: G06F 18/253 of extracted features
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/086 using evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: G06V 10/42 Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation

Abstract

A Self-Designing Intelligent Signal Processing System Capable of Evolutional Learning for Classification/Recognition of One and Multidimensional Signals is described which classifies data by an evolutionary learning environment that develops the features and algorithms that are best suited for the recognition problem under consideration. The System adaptively learns what data need to be extracted in order to recognize the given pattern with the least amount of processing. The System decides what features need to be selected for classification and/or recognition to fit a certain structure that leads to the least amount of processing according to the nature of the given data. The System disclosed herein is capable of recognizing an enormously large number of patterns with a high accuracy.

Claims

1. A self-designing intelligent signal processing system, comprising:

(a) means for receiving signals corresponding to plural images, wherein each of the plural images are not related to other ones of the plural images;

(b) means for extracting coefficients from the received signals in more than one transform domain, wherein the coefficients correspond to the plural images;

(c) means for determining corresponding criteria in each of the more than one transform domain using the extracted coefficients;

(d) an adaptive learning means that determines a best characteristics and corresponding criteria to be extracted by each of the more than one transform domain in order to recognize the plural images with the least amount of processing;

(e) means for classifying the plural images from each more than one transform domain into selected groups according to the criteria corresponding to that one of the more than one transform domain over a range of adaptable parameters; and

(f) means for recognizing each of the plural images from the selected groups from the more than one transform domain, wherein the self-designing intelligent signal processing system is capable of recognizing a large number of different images with a high level of accuracy.

2. The self-designing intelligent signal processing system of claim 1, wherein the more than one transform domain comprises:

a discrete cosine transform (DCT);

HAAR transform domain; and

a singular values (SV) decomposition transform domain which cooperate to provide an output of digitized images, wherein said DCT, HAAR and SV transforms are arranged in parallel.

3. The self-designing intelligent signal processing system of claim 1, further comprising:

means for selecting an optimum criteria for use in classification of the plural images from noisy signals.

4. The self-designing intelligent signal processing method, comprising the steps of:

(a) receiving signals from a source, the signals including plural unrelated images;

(b) recognizing a pattern from the received signals using plural transforms domains to select from each one of the plural transform domains unique features and corresponding criteria;

(c) classifying the unique features into selected groups for each of the plural transform domains based on the corresponding criteria;

(d) determining fidelity of said selected groups; and

(e) updating each of the selected groups from said plural transform domains by an evolutionary recognition classifier that selects a different optimum feature and corresponding criteria for use by each of the plural transforms domains, whereby the fidelity of said selected groups is constantly improved over time based on the receiving signals to recognize each of the plural unrelated images with a high level of accuracy.

5. The self-designing intelligent signal processing method of claim 4, wherein the signals include:

one of a one-dimensional signals and a multi-dimensional signal.

6. The self-designing intelligent signal processing method of claim 4, wherein the patterns include digitized images.

7. The self-designing intelligent signal processing method of claim 4, wherein the patterns are developed by a back propagation of respective neural networks of each one of the plural transform domains.

8. The self-designing intelligent signal processing method of claim 4, wherein the patterns include noise signals.

9. The self-designing intelligent signal processing method of claim 4, further comprising the step of:

filtering out noise from the received signals.

10. The self-designing intelligent signal processing method of claim 4, wherein the patterns are clustered into digitized groups of common facial characteristics.

11. A multi-criteria multi-transform neural network (MCMTNN) classifier for image classification comprising:

(a) an input device for receiving input signals having plural different digitized images;

(b) plural transform domains in parallel for receiving the input signals from the input device, each of the plural transform domains comprising;

a pattern recognizer for receiving the input signals and developing transform coefficients for a particular domain and extracting plural features from the transform coefficients;

a criteria selector for selecting classification criteria from the transform coefficients for the particular domain; and

a neural network for classifying the plural digitized images into plural groups according to the selected criteria over a range of parameters; and

wherein the MCMTNN classifier adaptively determines the best features and criteria for use in each of the plural transform domains for the recognition problem under consideration with the least amount of processing to recognize the plural different digitized images.