A SELF-LEARNING SYSTEM AND METHODS FOR AUTOMATIC DOCUMENT RECOGNITION, AUTHENTICATION, AND INFORMATION EXTRACTION

• US 20170132866A1
• Filed: 06/19/2015
• Published: 05/11/2017
• Est. Priority Date: 06/20/2014
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Kuklinski, Theodore, Monk, Bruce
• Granted Patent: US 10,354,472 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06F 18/24765 Rule-based classification
• CPC Class Codes: G06F 21/44 Program or device authentication
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06V 30/413 Classification of content, e.g. text, photographs or tables
• CPC Class Codes: G06V 30/414 Extracting the geometrical structure, e.g. layout tree; Block segmentation, e.g. bounding boxes for graphics or text
• CPC Class Codes: G07D 7/202 using pattern matching
• CPC Class Codes: G07D 7/2083 Learning

Abstract

A computerized system for classifying and authenticating documents is provided. The Classification process involves the creation of a Unique Pair Feature Vector which provides the best discrimination information for each pair of Document Classes at every node in a Pairwise Comparison Nodal Network. The Nodal Network has a plurality of nodes, each node corresponding to the best discrimination information between two potential document classes. By performing a pairwise comparison of the potential documents using this nodal network, the document is classified. After classification, the document can be authenticated for validity.

Claims

1. ) A computerized method for developing a pairwise comparison nodal network for classification of an item capable of authentication comprising the steps of:

receiving electronic representations of a plurality of items capable of authentication of a first item class, a plurality of items capable of authentication of a second item class, and a plurality of items capable of authentication of a third item class, by at least one computer;

selecting, by the at least one computer, a plurality of regions of the scanned representation of the plurality of items, the plurality of regions being the same for each item of each of the first, second, and third item class;

recording, by the at least one computer, at least one measurement for each of the plurality of regions relating to the scanned representation of each of the plurality of items in each of the first, second, and third item class;

compiling and storing the measurements of the plurality of items of the first item class as a first feature vector by the at least one computer to a computerized storage;

compiling and storing the measurements of the plurality of items of the second item class as a second feature vector by the at least one computer to the computerized storage;

compiling and storing the measurements of the plurality of items of the third item class as a third feature vector by the at least one computer to the computerized storage;

comparing, by the at least one computer, the first feature vector with the second feature vector, the comparing step comprising identifying a plurality of the measurements having the greatest difference between the first feature vector and second feature vector, the identified plurality of measurements being the best distinguishing measurements between the first item class and second item class;

compiling and storing the identified plurality of the measurements having the greatest difference between the first feature vector and second feature vector as a first inter class feature vector by the at least one computer to the computerized storage;

comparing, by the at least one computer, the first feature vector with the third feature vector, the comparing step comprising identifying a second plurality of the measurements having the greatest difference between the first feature vector and third feature vector;

compiling and storing the second plurality of the measurements having the greatest difference between the first feature vector and third feature vector as a second inter class feature vector by the at least one computer to the computerized storage;

comparing, by the at least one computer, the second feature vector with the third feature vector, the comparing step comprising identifying a third plurality of the measurements having the greatest difference between the second feature vector and third feature vector;

compiling and storing the third plurality of the measurements having the greatest difference between the second feature vector and third feature vector as a third inter class feature vector by the at least one computer to the computerized storage;

forming, by the at least one computer, a pairwise comparison nodal network having a first comparison node comprising the first interclass feature vector, a second comparison node comprising the second interclass feature vector, and a third comparison node comprising the third interclass feature vector, wherein the pairwise comparison nodal network is usable by the at least one computer to compare interclass feature vector measurements of each node in a pairwise fashion, a result of an analysis by the pairwise comparison nodal network providing an identification of what class a candidate item being classified best corresponds to.

2. ) The method of claim 1 further comprising the step of identifying, the plurality of measurements of each of the first, second and third feature vector having a variance between each of the plurality of items greater than a predetermined level by the at least one computer.

3. ) The method of claim 1 wherein the plurality of items of the first, second, and third item classes is each an identification document.

4. ) The method of claim 1 wherein the pairwise comparison nodal network further comprises a plurality of nodes in addition to the first, second, and third node, each of the plurality of nodes corresponding to an interclass feature vector of a plurality of different item classes.

5. ) The method of claim 4 further comprising the step of ordering, by the at least one computer, the first, second and third nodes and the plurality of nodes such that the inter class feature vector of two of the most likely items classes is an initial node analyzed by the computerized pairwise comparison nodal network.

6. ) The method of claim 5 wherein the ordering is performed by the at least one computer based on a location-based proximity to a projected area of operation, wherein the area of operation is one of a geographic, electronic, network, node, or cloud location.

7. ) The method of claim 1 wherein the at least one measurement taken by the at least one computer are measurements of at least one of color, shape identification, luminance, statistical properties, line content, brightness, shading, and spectral measurements.

8. ) A computerized method for classifying and authenticating an item capable of authentication using a pairwise comparison nodal network comprising the steps of:

obtaining an electronic representation of a candidate item capable of authentication using a computerized input device into at least one computer;

selecting, by the at least one computer, a predetermined plurality of regions of the scanned candidate item;

recording, by the at least one computer, at least one measurement for each of the plurality of regions;

classifying the candidate item as one of a plurality of item classes using a pairwise comparison nodal network of the at least one computer, the pairwise comparison nodal network having a first node comprising a plurality of measurements that best distinguish between a first item class and a second item class, a second node comprising a plurality of measurements that best distinguish between a first item class and a third item class, and a third node comprising a plurality of measurements that best distinguish between a second item class and a third item class, the classifying being performed in a pairwise comparison fashion comprising;

analyzing the measurements of the candidate item at the first node by the at least one computer, the analyzing resulting in a computerized selection that the candidate item better corresponds to either the first item class or the second item class;

wherein if the computerized selection corresponds to the first item class, analyzing the measurements of the candidate item at the second node, resulting in a computerized selection that the candidate item better corresponds to either the first item class or the third item class, or wherein if the computerized selection corresponds to the second item class, analyzing the measurements of the candidate item at the third node, resulting in a computerized selection that the candidate item better corresponds to either the second item class or the third item class, wherein the analysis of the first node and one of the second or the third node results in a classification of the candidate item; and

authenticating, by the at least one computer, the classified candidate item after the classification analyzing steps.

9. ) The method of claim 8 wherein the candidate item is an identification document.

10. ) The method of claim 8 wherein the first item class and second item class of the first node are selected to be the most and second most likely item to be scanned.

11. ) The method of claim 8 wherein the first item class and second item class of the first node are selected to be the most and second most geographically proximate State issued identification documents.

12. ) The method of claim 8 wherein the authenticating step comprises performing optical character recognition of identification information on the item by the at least one computer.

13. ) The method of claim 8 further comprising a fourth node comprising a plurality of measurements that best distinguish between the first class of items and a class of known fake items, and further comprising the step of analyzing, by the at least one computer, the fourth node, resulting in a computerized selection that the candidate item corresponds to the first item class or the class of known fake items.

14. ) The method of claim 8 wherein the at least one measurement taken by the at least one computer are measurements of at least one of color, shape identification, luminance, statistical properties, line content, brightness, shading, and spectral measurements.

15. ) The method of claim 8 wherein the at least one computer is a mobile wireless computing device, and wherein the computerized input is a camera.

16. ) The method of claim 15 further comprising the step of determining, by the at least one computer, if a classification confidence based on the classifying step is above a predetermined level, the classification confidence being calculated based on a plurality of validation parameters;

and

determining, based on the classification confidence, that the document is real if the classification is above the predetermined level, or that the document is one of real but poor quality, damaged, forged, altered, a document subclass, and an unknown document if the classification confidence is below a predetermined level.

17. ) The method of claim 8 further comprising the step of calculating, based on the classifying and authenticating steps by the at least one computer, a confidence of validity, and comprising the step of displaying the confidence of validity on a display of the at least one computer.

18. ) A computerized method for classifying and authenticating a document using a pairwise comparison nodal network comprising the steps of:

obtaining an electronic representation of a candidate document using a computerized input device into at least one computer;

selecting, by the at least one computer, a predetermined plurality of regions of the scanned candidate document;

recording, by the at least one computer, at least one measurement for each of the plurality of regions wherein the at least one measurement taken by the at least one computer are at least one of color, shape identification, luminance, line content, brightness, shading, and ultraviolet measurements;

classifying the candidate document as one of a first plurality of document classes using a pairwise comparison nodal network of the at least one computer, the pairwise comparison nodal network having a first node comprising a plurality of measurements that best distinguish between a first document class and a second document class wherein the first document class and second document class of the first node are selected to be the most and second most likely documents to be scanned, a second node comprising a plurality of measurements that best distinguish between a first document class and a third document class, and a third node comprising a plurality of measurements that best distinguish between a second document class and a third document class, the classifying being performed in a pairwise comparison fashion comprising;

analyzing the measurements of the candidate document at the first node by the at least one computer, the analyzing resulting in a computerized selection that the candidate document better corresponds to either the first document class or the second document class;

wherein if the computerized selection corresponds to the first document class, analyzing the measurements of the candidate document at the second node, resulting in a computerized selection that the candidate document better corresponds to either the first document class or the third document class, or wherein if the computerized selection corresponds to the second document class, analyzing the measurements of the candidate document at the third node, resulting in a computerized selection that the candidate document better corresponds to either the second document class or the third document class, wherein the analysis of the first node and one of the second or the third node results in a preliminary classification of the candidate document;

classifying the candidate document as one of a second plurality of document classes, wherein the second plurality of document classes are identification documents, the second plurality of document classes being a subclass of the one of the first plurality of document classes, the classifying comprising using a second pairwise comparison nodal network having a plurality of nodes, each node corresponding to a plurality of measurements that best distinguish between two of the second plurality of document classes, wherein the analysis by the second pairwise comparison nodal network results in a classification of the preliminarily classified document;

authenticating, by the at least one computer, the classified candidate document after the classification analyzing; and

calculating, based on the classifying and authenticating steps by the at least one computer, a confidence of validity, and comprising the step of displaying the confidence of validity on a display of the at least one computer.

19. ) The method of claim 18 wherein the at least one computer is a mobile wireless computing device, and wherein the computerized input is a camera.

20. ) The method of claim 18 further comprising the step of determining, by the at least one computer, if a classification confidence based on the classifying step is above a predetermined level, the classification confidence being calculated based on a plurality of validation parameters;

and

determining, based on the classification confidence, that the document is real if the classification is above the predetermined level, or that the document is one of real but poor quality, damaged, forged, altered, a document subclass, and an unknown document if the classification confidence is below a predetermined level.