Cognitive memory and auto-associative neural network based search engine for computer and network located images and photographs

• US 7,333,963 B2
• Filed: 10/07/2005
• Issued: 02/19/2008
• Est. Priority Date: 10/07/2004
• Status: Expired due to Fees

• Current Assignee: Bernard Widrow, Brian Mitchell Percival, Juan Carlos Aragon
• Original Assignee: Bernard Widrow, Brian Mitchell Percival, Juan Carlos Aragon
• Inventors: Aragon, Juan Carlos, Percival, Brian Mitchell, Widrow, Bernard
• Primary Examiner(s): Vincent; David
• Assistant Examiner(s): BUSS, BENJAMIN J
• Application Number:
  US11/245,860
• Publication Number:
  US 20060133699A1
• Time in Patent Office: 865 Days
• Field of Search: 706/15, 706/16, 706/18, 706/20, 706/25, 706/934, 382/155, 382/159
• US Class Current: 706/18
• CPC Class Codes: G06F 16/583 using metadata automatically derived from the content
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G06V 10/70 using pattern recognition or machine learning optical pattern recognition or electronic computations therefor G06V10/88
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: Y10S 706/934 Information retrieval or Information management

Abstract

Designs for cognitive memory systems storing input data, images, or patterns, and retrieving it without knowledge of where stored when cognitive memory is prompted by query pattern that is related to sought stored pattern. Retrieval system of cognitive memory uses autoassociative neural networks and techniques for pre-processing query pattern to establish relationship between query pattern and sought stored pattern, to locate sought pattern, and to retrieve it and ancillary data. Cognitive memory, when connected to computer or information appliance introduces computational architecture that applies to systems and methods for navigation, location and recognition of objects in images, character recognition, facial recognition, medical analysis and diagnosis, video image analysis, and to photographic search engines that when prompted with a query photograph containing faces and objects will retrieve related photographs stored in computer or other information appliance, and will identify URL'"'"'s of related photographs and documents stored on the World Wide Web.

Claims

1. A search engine for searching a computer or other information appliance, wherein said search engine seeks stored images, said stored images depicting persons'"'"' faces or objects of interest, wherein said stored images are retrieved in response to receipt of a query comprising a query image;

said query image depicting one or more persons'"'"' faces, or depicting one or more objects of interest;

wherein said search engine comprises;

(a) first means for locating images of persons'"'"' faces or objects of interest within both said query image and said stored images, said first means comprising;

i. first autoassociative neural network trained on first low resolution input patterns and first variations, wherein each of said first low resolution input patterns depicts one person'"'"'s face or one object of interest, wherein each of said first low resolution input patterns and said first variations contains 2000 or fewer pixels, and wherein said first variations are created from said first low resolution input patterns by at least one of or any combination of rotation, translation, changes in scale, brightness, and contrast, and other image processing techniques;

ii. first window means for scanning over said query image and over all said stored images, creating second low resolution input patterns and second variations, said second variations generated by at least one of or any combination of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

iii. means for applying, as inputs to trained said first autoassociative neural network, said second low resolution input patterns and said second variations, seeking a low error in the difference between the input and output of said first autoassociative neural network, where said low error in difference, when below a preset threshold, indicates a detected face or object of interest;

(b) second means for interrelating said query image, depicting persons'"'"' faces or objects of interest, to stored images depicting the same persons'"'"' faces or objects of interest, said second means comprising;

second autoassociative neural network trained on first high resolution input patterns and first high resolution variations, wherein each of said first high resolution input patterns depicts one detected person'"'"'s face or object of interest, said detected person'"'"'s face or object of interest detected in said query image by said first means, wherein each of said first high resolution input patterns and said first high resolution variations contains 2000 or more pixels, and wherein said first high resolution variations are created from said first high resolution input patterns by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

ii. second window means for scanning over said stored images, creating second high resolution input patterns and second high resolution variations, said second high resolution variations generated by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, and other image processing techniques;

iii. means for applying, as inputs to trained said second autoassociative neural network, said second high resolution input patterns and said second high resolution variations derived from a given stored image by said second window means, seeking a second low error in the difference between the input and output of said second autoassociative neural network; and

iv. means for identifying the given stored image that is related to said query image, wherein said given stored image is related to said query image when said second low error is below a pre-set threshold; and

(c) third means for delivering as output response to said query image said stored images related by said second means.

2. In a computer or other information appliance, a computer-implemented method for searching and interrelating stored original images, said stored original images depicting persons'"'"' faces or objects of interest, said method comprising:

(a) assigning a unique identifier for each stored original image;

(b) locating and segmenting images of faces or objects of interest depicted in each said stored original image;

(c) creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) creating separate folders for storage of each segmented image of a face or object of interest, each of said folders further containing said variations generated from each said segmented image of a face or object of interest, each of said folders further containing said unique identifier of said stored original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) training an individual autoassociative neural network for each of said folders, the training patterns for said individual autoassociative neural network comprising the images and their respective variations stored in the respective folder, thus relating each said individual autoassociative neural network to the folder containing its training patterns;

(f) storing the weights and parameters of each trained said individual autoassociative neural network;

(g) applying input patterns to each trained autoassociative neural network, said input patterns comprising all images stored in all said folders;

(h) determining the error, the difference between a given input pattern and the corresponding output pattern of a particular trained autoassociative neural network;

(i) adding all unique identifiers stored in the folder of said given input pattern to the list of unique identifiers stored in the folder related to said particular trained autoassociative neural network, if said error is below a pre-set threshold; and

(j) performing steps (h) and (i) for all pairings of input patterns and trained autoassociative neural networks;

whereby each of said folders contains a unique identifier or a set of unique identifiers, thereby connecting the images of faces or objects of interest in each folder to a stored original image or to a set of stored original images, and thereby interrelating stored original images that depict faces or objects of interest in common.

3. The method of claim 2, further comprising:

(a) grouping a particular folder with other folders when images stored in said other folders are applied as input patterns to the autoassociative neural network related to said particular folder and the error between input and output patterns of said autoassociative neural network is below a pre-set threshold;

(b) performing said grouping step for each folder and establishing all possible such groups, each such group containing all of the folders containing images of a given face or object of interest;

(c) attaching the name or keywords to a group if one or more of the folders of the group contain images that were derived from a stored original image in which segmented faces or objects of interest had been identified by said name or keywords; and

(d) attaching the name or keywords to a group if one or more of the images stored in the folders of the group were manually identified by said name or keywords.

4. The method of claim 2, further comprising:

(a) locating and segmenting images of faces or objects of interest depicted in a query image;

(b) creating query variations from each of the segmented images of faces or objects of interest depicted in said query image, said query variations created by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(c) applying query input patterns to a given trained autoassociative neural network related to a given folder, said query input patterns comprising all segmented images of faces and objects of interest in the query image, said query input patterns further comprising all said query variations;

(d) determining the query error, the difference between a given query input pattern and the resulting output pattern from a given trained autoassociative neural network;

(e) delivering as part of the output response to the query image the stored original images corresponding to all unique identifiers stored in the folder corresponding to said given trained autoassociative network if said query error is below a pre-set threshold; and

(f) performing steps (c), (d), and (e) for said trained autoassociative neural networks corresponding to all said folders.

5. The method of claim 3, further comprising:

(a) identifying all matching groups, said all matching groups being all said groups with attached matching names or matching keywords, said matching names or matching keywords being those names or keywords that match the text of a text query;

(b) identifying all matching folders, said all matching folders being all folders of said all matching groups;

(c) identifying all matching unique identifiers, said all matching unique identifiers being all unique identifiers stored in said all matching folders; and

(d) delivering as output the original stored images corresponding to said all matching unique identifiers.

6. In a computer or other information appliance, a computer-implemented method for searching and interrelating stored original images, said stored original images depicting persons'"'"' faces or objects of interest, said method comprising:

(a) locating each stored original image;

(b) locating and segmenting images of faces or objects of interest depicted in each said stored original image;

(c) creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) creating separate folders for storage of each image of a segmented face or object of interest, each of said folders further containing said variations generated from each said segmented image of a face or object of interest, wherein a folder represents any means of associating the contents thereof;

(e) training an individual autoassociative neural network for each of said folders, the training patterns for said individual autoassociative neural network comprising said segmented images of faces or objects of interest, and further comprising their respective said variations stored in the same folder;

(f) storing the weights and parameters of said individual autoassociative neural network;

(g) applying input patterns to each trained autoassociative neural network, said input patterns comprising all images stored in all said folders;

(h) grouping together the folder containing a given input pattern and the folder associated with a particular trained autoassociative neural network, whenever the error is below a pre-set threshold, wherein said error is the difference between said given input pattern and said particular trained autoassociative neural network output pattern;

(i) performing said grouping step for each folder and establishing all possible such groups, each such group containing all of the folders containing images of a given face or object of interest;

(j) attaching the name or keywords to a group if one or more of the folders of the group contain images that were derived from a stored original image in which segmented faces or objects of interest had been identified by said name or keywords; and

(k) attaching the name or keywords to a group if one or more of the images stored in the folders of the group were manually identified by said name or keywords.

7. The method of claim 6, further comprising:

(a) locating and segmenting images of faces or objects of interest depicted in a query image;

(b) creating query variations from each of the segmented images of faces or objects of interest depicted in the query image, said query variations created by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(c) applying query input patterns to a given trained autoassociative neural network associated with a given folder, said query input patterns comprising all segmented images of faces or objects of interest in the query image, said query input patterns further comprising all said query variations;

(d) determining the query error, said query error being the difference between a given query input pattern and the resulting output pattern from a given trained autoassociative neural network;

(e) delivering as part of the output response to the query the name or keywords attached to the group containing the folder corresponding to said given trained autoassociative neural network if said query error is below a pre-set threshold; and

(f) performing steps (c), (d), and (e) for said trained autoassociative neural networks corresponding to all said folders.

8. In a computer or other information appliance, a computer-implemented method for searching and interrelating original images, said original images depicting persons'"'"' faces or objects of interest, wherein said original images are stored on web servers or other information storage appliances, wherein said original images are accessible to said computer or other information appliance, said method comprising:

(a) crawling or searching a computer network, the World Wide Web, or other types of interconnected networks, to collect original images and their universal resource locators, URLs, said URLs indicating the locations of said collected original images;

(b) locating and segmenting images of faces or objects of interest depicted in collected original images;

(c) creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) creating a single folder for storage of said each segmented image of a face or an object of interest, each said single folder further containing said variations generated from each said segmented image of a face or an object of interest, each said single folder further containing the URL of said collected original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) applying input patterns to an autoassociative neural network, said input patterns comprising the images contained within said single folder;

(f) determining the error, wherein said error is the difference between one of said input patterns from one said single folder and the corresponding output pattern from said autoassociative neural network;

(g) training said autoassociative neural network with the patterns contained in said single folder, if said error is above a pre-set threshold for every pattern in said single folder;

(h) performing the following steps for a hit pattern in said single folder, wherein said hit pattern is a pattern for which said error is below said pre-set threshold;

i. comparing said hit pattern to all patterns in all other folders; and

ii. adding the hit URL to the folder containing the matching pattern, wherein said matching pattern is the pattern having the smallest difference between said matching pattern and said hit pattern, wherein said hit URL is the URL in the folder containing said hit pattern; and

(i) performing steps (d), (e), (f), (g), and (h) for each said segmented image;

whereby each folder contains a URL or a set of URLs, thereby connecting the images of faces or objects of interest in each folder to an original image or to a set of original images, thus interrelating original images depicting faces or objects of interest that are in common.

9. The method of claim 8, further comprising:

(a) locating and segmenting images of faces or objects of interest depicted in a query image;

(b) creating query variations from each of the segmented images of faces or objects of interest depicted in said query image, said query variations created by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(c) applying query input patterns to said autoassociative neural network, said query input patterns comprising all segmented images of faces or objects of interest in the query image, said query input patterns further comprising all said query variations;

(d) determining the query error for each query input pattern, wherein said query error is the difference between a given query input pattern and the resulting output pattern from said autoassociative neural network; and

(e) performing the following steps for each query hit pattern, wherein a query hit pattern is a query input pattern resulting in a query error below a pre-set threshold;

i. comparing said query hit pattern to all patterns in each said folder; and

ii. delivering as part of the output response to the query image the one or more URLs contained in the folder containing the query matching pattern, wherein said query matching pattern is the pattern having the smallest difference between said query matching pattern and said query hit pattern.

10. In a computer or other information appliance, a computer-implemented method for searching and interrelating original images, said original images depicting persons faces or objects of interest, wherein said original images are stored on web servers or other information storage appliances, wherein said original images are accessible to said computer or other information appliance, said method comprising:

(a) crawling or searching a computer network, the World Wide Web, or other types of interconnected networks, to collect original images and their universal resource locators, URLs, said URLs indicating the locations of said collected original images;

(b) locating and segmenting images of faces or objects of interest depicted in collected original images;

(c) creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) creating a single folder for storage of said each segmented image of a face or an object of interest, each said single folder further containing said variations generated from each said segmented image of a face or an object of interest, each said single folder further containing the URL of said collected original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) applying input patterns to each autoassociative neural network of a plurality of autoassociative neural networks, said input patterns comprising the images contained within said single folder;

(f) determining the errors, wherein said errors are the differences between one of said input patterns from one said single folder and the corresponding output pattern from said plurality of autoassociative neural networks;

(g) training one of said autoassociative neural networks with the patterns contained in said single folder thereby associating said single folder with said one of said autoassociative neural networks, if all said errors are above a pre-set threshold for every pattern in said single folder;

(h) performing the following steps for a hit pattern in said single folder, wherein said hit pattern is a pattern for which said error is below said pre-set threshold;

i. comparing said hit pattern to all patterns in all folders associated with any of said autoassociative neural networks whose error, when presented with said hit pattern, is below said pre-set threshold; and

ii. adding the hit URL to the folder containing the matching pattern, wherein said matching pattern is the pattern having the smallest difference between said matching pattern and said hit pattern, wherein said hit URL is the URL in the folder containing said hit pattern; and

(i) performing steps (d), (e), (f), (g), and (h) for each said segmented image;

whereby each folder contains a URL or a set of URLs, thereby connecting the images of faces or objects of interest in each folder to an original image or to a set of original images, thus interrelating original images depicting faces or objects of interest that are in common.

11. The method of claim 10, further comprising:

(a) locating and segmenting images of faces or objects of interest depicted in a query image;

(b) creating query variations from each of the segmented images of faces or objects of interest depicted in said query image, said query variations created by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(c) applying query input patterns to said each autoassociative neural network of a plurality of autoassociative neural networks, said query input patterns comprising all segmented images of faces or objects of interest in the query image, said query input patterns further comprising all said query variations;

(d) determining the query errors for each query input pattern, wherein said query errors are the differences between a given query input pattern and the resulting output patterns from all said autoassociative neural networks; and

(e) performing the following steps for each query hit pattern, wherein a query hit pattern is a query input pattern resulting in a query error below a query pre-set threshold;

i. comparing said query hit pattern to all patterns in all folders associated with any of said autoassociative neural networks whose query error, when presented with said query hit pattern, is below said query pre-set threshold; and

ii. delivering as part of the output response to the query image the one or more URLs contained in the folder containing the query matching pattern, wherein said query matching pattern is the pattern having the smallest difference between said query matching pattern and said query hit pattern.

12. The method of claim 11, further comprising:

(a) compiling a list of all image URLs, wherein said image URLs are URLs delivered as part of the output response to said input query image, wherein said input query image is the image component of an input query;

(b) compiling a list of all text URLs, wherein said text URLs are URLs delivered as the output response to a keyword search of a computer, a network of computers, the World Wide Web, or other interconnected networks of computers, wherein said keyword search is performed in response to the keyword component of said input query; and

(c) delivering as the output to said input query all URLs that are in common to said list of all image URLs and to said list of all text URLs.

13. A computer program product for use with a computer or information appliance comprising:

a computer operable medium having computer readable code, the computer readable code being effective to perform a computer-implemented method for searching and interrelating stored original images, said stored original images depicting persons'"'"' faces or objects of interest, said computer-implemented method comprising;

(a) assigning a unique identifier for each stored original image;

(b) locating and segmenting images of faces or objects of interest depicted in each said stored original image;

(c) creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) creating separate folders for storage of each segmented image of a face or object of interest, each of said folders further containing said variations generated from each said segmented image of a face or object of interest, each of said folders further containing said unique identifier of said stored original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) training an individual autoassociative neural network for each of said folders, the training patterns for said individual autoassociative neural network comprising the images and their respective variations stored in the respective folder, thus relating each said individual autoassociative neural network to the folder containing its training patterns;

(f) storing the weights and parameters of each trained said individual autoassociative neural network;

(g) applying input patterns to each trained autoassociative neural network, said input patterns comprising all images stored in all said folders;

(h) determining the error, the difference between a given input pattern and the corresponding output pattern of a particular trained autoassociative neural network;

(i) adding all unique identifiers stored in the folder of said given input pattern to the list of unique identifiers stored in the folder related to said particular trained autoassociative neural network, if said error is below a pre-set threshold; and

(j) performing steps (h) and (i) for all pairings of input patterns and trained autoassociative neural networks;

whereby each of said folders contains a unique identifier or a set of unique identifiers, thereby connecting the images of faces or objects of interest in each folder to a stored original image or to a set of stored original images, and thereby interrelating stored original images that depict faces or objects of interest in common.

14. A computer system for searching and interrelating stored images, said stored images depicting persons'"'"' faces or objects of interest, said computer system comprising:

(a) means for assigning a unique identifier for each stored original image;

(b) means for locating and segmenting images of faces or objects of interest depicted in each said stored original image;

(c) means for creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) means for creating separate folders for storage of each segmented image of a face or object of interest, each of said folders further containing said variations generated from each said segmented image of a face or object of interest, each of said folders further containing said unique identifier of said stored original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) means for training an individual autoassociative neural network for each of said folders, the training patterns for said individual autoassociative neural network comprising the images and their respective variations stored in the respective folder, thus relating each said individual autoassociative neural network to the folder containing its training patterns;

(f) means for storing the weights and parameters of each trained said individual autoassociative neural network;

(g) means for applying input patterns to each trained autoassociative neural network, said input patterns comprising all images stored in all said folders;

(h) means for determining the error, the difference between a given input pattern and the corresponding output pattern of a particular trained autoassociative neural network;

(i) means for adding all unique identifiers stored in the folder of said given input pattern to the list of unique identifiers stored in the folder related to said particular trained autoassociative neural network, if said error is below a pre-set threshold; and

(j) means for performing steps (h) and (i) for all pairings of input patterns and trained autoassociative neural networks;

whereby each of said folders contains a unique identifier or a set of unique identifiers, thereby connecting the images of faces or objects of interest in each folder to a stored original image or to a set of stored original images, and thereby interrelating stored original images that depict faces or objects of interest in common.

15. A computer program product for use with a computer or information appliance comprising:

a computer operable medium having computer readable code, the computer readable code being effective to perform a computer-implemented method for searching and interrelating stored original images, said stored original images depicting persons faces or objects of interest, said method comprising;

(a) locating each stored original image;

(b) locating and segmenting images of faces or objects of interest depicted in each said stored original image;

(c) creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) creating separate folders for storage of each image of a segmented face or object of interest, each of said folders further containing said variations generated from each said segmented image of a face or object of interest, wherein a folder represents any means of associating the contents thereof;

(e) training an individual autoassociative neural network for each of said folders, the training patterns for said individual autoassociative neural network comprising said segmented images of faces or objects of interest, and further comprising their respective said variations stored in the same folder;

(f) storing the weights and parameters of said individual autoassociative neural network;

(g) applying input patterns to each trained autoassociative neural network, said input patterns comprising all images stored in all said folders;

(h) grouping together the folder containing a given input pattern and the folder associated with a particular trained autoassociative neural network, whenever the error is below a pre-set threshold, wherein said error is the difference between said given input pattern and said particular trained autoassociative neural network output pattern;

(i) performing said grouping step for each folder and establishing all possible such groups, each such group containing all of the folders containing images of a given face or object of interest;

(j) attaching the name or keywords to a group if one or more of the folders of the group contain images that were derived from a stored original image in which segmented faces or objects of interest had been identified by said name or keywords; and

(k) attaching the name or keywords to a group if one or more of the images stored in the folders of the group were manually identified by said name or keywords.

16. A computer system for searching and interrelating stored original images, said stored original images depicting persons faces or objects of interest, said system comprising:

(a) means for locating each stored original image;

(b) means for locating and segmenting images of faces or objects of interest depicted in each said stored original image;

(c) means for creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) means for creating separate folders for storage of each image of a segmented face or object of interest, each of said folders further containing said variations generated from each said segmented image of a face or object of interest, wherein a folder represents any means of associating the contents thereof;

(e) means for training an individual autoassociative neural network for each of said folders, the training patterns for said individual autoassociative neural network comprising said segmented images of faces or objects of interest, and further comprising their respective said variations stored in the same folder;

(f) means for storing the weights and parameters of said individual autoassociative neural network;

(g) means for applying input patterns to each trained autoassociative neural network, said input patterns comprising all images stored in all said folders;

(h) means for grouping together the folder containing a given input pattern and the folder associated with a particular trained autoassociative neural network, whenever the error is below a pre-set threshold, wherein said error is the difference between said given input pattern and said particular trained autoassociative neural network output pattern;

(i) means for performing said grouping step for each folder and establishing all possible such groups, each such group containing all of the folders containing images of a given face or object of interest;

(j) means for attaching the name or keywords to a group if one or more of the folders of the group contain images that were derived from a stored original image in which segmented faces or objects of interest had been identified by said name or keywords; and

(k) means for attaching the name or keywords to a group if one or more of the images stored in the folders of the group were manually identified by said name or keywords.

17. A computer program product for use with a computer or information appliance comprising:

a computer operable medium having computer readable code, the computer readable code being effective to perform a computer-implemented method for searching and interrelating original images, said original images depicting persons'"'"' faces or objects of interest, wherein said original images are stored on web servers or other information storage appliances, wherein said original images are accessible to said computer or other information appliance, said method comprising;

(a) crawling or searching a computer network, the World Wide Web, or other types of interconnected networks, to collect original images and their universal resource locators, URLs, said URLs indicating the locations of said collected original images;

(b) locating and segmenting images of faces or objects of interest depicted in collected original images;

(c) creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) creating a single folder for storage of said each segmented image of a face or an object of interest, each said single folder further containing said variations generated from each said segmented image of a face or an object of interest, each said single folder further containing the URL of said collected original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) applying input patterns to an autoassociative neural network, said input patterns comprising the images contained within said single folder;

(f) determining the error, wherein said error is the difference between one of said input patterns from one said single folder and the corresponding output pattern from said autoassociative neural network;

(g) training said autoassociative neural network with the patterns contained in said single folder, if said error is above a pre-set threshold for every pattern in said single folder;

(h) performing the following steps for a hit pattern in said single folder, wherein said hit pattern is a pattern for which said error is below said pre-set threshold;

i. comparing said hit pattern to all patterns in all other folders; and

ii. adding the hit URL to the folder containing the matching pattern, wherein said matching pattern is the pattern having the smallest difference between said matching pattern and said hit pattern, wherein said hit URL is the URL in the folder containing said hit pattern; and

(i) performing steps (d), (e), (f), (g), and (h) for each said segmented image;

whereby each folder contains a URL or a set of URLs, thereby connecting the images of faces or objects of interest in each folder to an original image or to a set of original images, thus interrelating original images depicting faces or objects of interest that are in common.

18. A computer system for searching and interrelating original images depicting persons'"'"' faces or objects of interest, wherein said original images are stored on external web servers or other information storage appliances, and wherein said original images are accessible to said computer system, said system comprising:

(a) means for crawling or searching a computer network, the World Wide Web, or other types of interconnected networks, to collect original images and their universal resource locators, URLs, said URLs indicating the locations of said collected original images;

(b) means for locating and segmenting images of faces or objects of interest depicted in collected original images;

(c) means for creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) means for creating a single folder for storage of said each segmented image of a face or an object of interest, each said single folder further containing said variations generated from each said segmented image of a face or an object of interest, each said single folder further containing the URL of said collected original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) means for applying input patterns to an autoassociative neural network, said input patterns comprising the images contained within said single folder;

(f) means for determining the error, wherein said error is the difference between one of said input patterns from one said single folder and the corresponding output pattern from said autoassociative neural network;

(g) means for training said autoassociative neural network with the patterns contained in said single folder, if said error is above a pre-set threshold for every pattern in said single folder;

(h) means for performing the following steps for a hit pattern in said single folder, wherein said hit pattern is a pattern for which said error is below said pre-set threshold;

i. comparing said hit pattern to all patterns in all other folders; and

ii. adding the hit URL to the folder containing the matching pattern, wherein said matching pattern is the pattern having the smallest difference between said matching pattern and said hit pattern, wherein said hit URL is the URL in the folder containing said hit pattern; and

(i) means for performing steps (d), (e), (f), (g), and (h) for each said segmented image;

whereby each folder contains a URL or a set of URLs, thereby connecting the images of faces or objects of interest in each folder to an original image or to a set of original images, thus interrelating original images depicting faces or objects of interest that are in common.

19. A computer program product for use with a computer or information appliance comprising:

a computer operable medium having computer readable code, the computer readable code being effective to perform a computer-implemented method for searching and interrelating original images, said original images depicting persons'"'"' faces or objects of interest, wherein said original images are stored on web servers or other information storage appliances, wherein said original images are accessible to said computer or other information appliance, said method comprising;

(a) crawling or searching a computer network, the World Wide Web, or other types of interconnected networks, to collect original images and their universal resource locators, URLs, said URLs indicating the locations of said collected original images;

(b) locating and segmenting images of faces or objects of interest depicted in collected original images;

(c) creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) creating a single folder for storage of said each segmented image of a face or an object of interest, each said single folder further containing said variations generated from each said segmented image of a face or an object of interest, each said single folder further containing the URL of said collected original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) applying input patterns to each autoassociative neural network of a plurality of autoassociative neural networks, said input patterns comprising the images contained within said single folder;

(f) determining the errors, wherein said errors are the differences between one of said input patterns from one said single folder and the corresponding output pattern from said plurality of autoassociative neural networks;

(g) training one of said autoassociative neural networks with the patterns contained in said single folder thereby associating said single folder with said one of said autoassociative neural networks, if all said errors are above a pre-set threshold for every pattern in said single folder;

(h) performing the following steps for a hit pattern in said single folder, wherein said hit pattern is a pattern for which said error is below said pre-set threshold;

i. comparing said hit pattern to all patterns in all folders associated with any of said autoassociative neural networks whose error, when presented with said hit pattern, is below said pre-set threshold; and

ii. adding the hit URL to the folder containing the matching pattern, wherein said matching pattern is the pattern having the smallest difference between said matching pattern and said hit pattern, wherein said hit URL is the URL in the folder containing said hit pattern; and

(i) performing steps (d), (e), (f), (g), and (h) for each said segmented image;

whereby each folder contains a URL or a set of URLs, thereby connecting the images of faces or objects of interest in each folder to an original image or to a set of original images, thus interrelating original images depicting faces or objects of interest that are in common.

20. A computer system for searching and interrelating original images depicting faces or objects of interest, wherein said original images are stored on web servers or other information storage appliances, and said original images are accessible to said computer system, said system further comprising:

(a) means for crawling or searching a computer network, the World Wide Web, or other types of interconnected networks, to collect original images and their universal resource locators, URLs, said URLs indicating the locations of said collected original images;

(b) means for locating and segmenting images of faces or objects of interest depicted in collected original images;

(c) means for creating variations of each segmented image of said faces or objects of interest by at least one of or any combinations of rotation, translation, changes in scale, brightness, and contrast, spatial filtering, frequency filtering, spatial frequency filtering, edge detection, perspective transformation, warping, distorting, distortion correction, image to image registration, gray-level histogram modification or equalization, adjusting color characteristics, varying or adjusting color saturation, removing color, distending, compressing, squeezing, shearing, and changes in intensity;

(d) means for creating a single folder for storage of said each segmented image of a face or an object of interest, each said single folder further containing said variations generated from each said segmented image of a face or an object of interest, each said single folder further containing the URL of said collected original image from which said segmented image of a face or object of interest was derived, wherein a folder represents any means of associating the contents thereof;

(e) means for applying input patterns to each autoassociative neural network of a plurality of autoassociative neural networks, said input patterns comprising the images contained within said single folder;

(f) means for determining the errors, wherein said errors are the differences between one of said input patterns from one said single folder and the corresponding output pattern from said plurality of autoassociative neural networks;

(g) means for training one of said autoassociative neural networks with the patterns contained in said single folder thereby associating said single folder with said one of said autoassociative neural networks, if all said errors are above a pre-set threshold for every pattern in said single folder;

(h) means for performing the following steps for a hit pattern in said single folder, wherein said hit pattern is a pattern for which said error is below said pre-set threshold;

i. comparing said hit pattern to all patterns in all folders associated with any of said autoassociative neural networks whose error, when presented with said hit pattern, is below said pre-set threshold; and

ii. adding the hit URL to the folder containing the matching pattern, wherein said matching pattern is the pattern having the smallest difference between said matching pattern and said hit pattern, wherein said hit URL is the URL in the folder containing said hit pattern; and

(i) means for performing steps (d), (e), (f), (g), and (h) for each said segmented image;

whereby each folder contains a URL or a set of URLs, thereby connecting the images of faces or objects of interest in each folder to an original image or to a set of original images, thus interrelating original images depicting faces or objects of interest that are in common.