Methods and apparatus for the automatic classification of patterns

• US 4,454,610 A
• Filed: 01/06/1982
• Issued: 06/12/1984
• Est. Priority Date: 05/19/1978
• Status: Expired due to Term

• Current Assignee: Transaction Sciences Corporation
• Original Assignee: Transaction Sciences Corporation
• Inventors: Sziklai, George C.
• Primary Examiner(s): Boudreau, Leo H.
• Application Number:
  US06/337,393
• Time in Patent Office: 888 Days
• Field of Search: 382/3, 382/33-34, 382/41-43, 179/1 SA, 235/457, 235/487, 235/489, 235/494, 235/470, 235/379-382, 340/825.3, 340/825.31, 340/825.34, 355/52
• US Class Current: 382/119
• CPC Class Codes: G07C 9/247 electronically, e.g. by comparing signal of hand-writing with a reference signal from the pass

Abstract

Methods and apparatus for the automatic classification of patterns are disclosed in which the intensity or gray level values of selected pixels of digital images of functions of position-invariant transforms of patterns to be classified, displayed as intensity functions, are correlated with one or more stored sets of values, each such stored set of values corresponding to one of the classes into which the unclassified patterns are to be classified. Methods of determining the coordinates of the selected pixels for use in automatic signature verfication and automatic handwritten numeral recognition are disclosed.

Claims

1. Apparatus for the classification of patterns, comprising:

unclassified representation generating means for generating a representation of at least part of a function of a position-invariant transform of each unclassified pattern presented thereto;

transform function representation element value signal set producing means for producing from each of said unclassified representations a set of element value signals each of which represents the value of a property of one of a predetermined constellation of elements thereof; and

correlating means for correlating each set of said transform function representation element value signals with at least one set of reference signals;

said elements of said unclassified representations being selected in accordance with a statistical property of a plurality of reference representations each of which represents at least part of said function of said position-invariant transform of a reference pattern, and each one of said plurality of reference patterns being a member of a subset of said plurality of reference patterns all of the members of which are cosignificative and also being a member of a subset of said plurality of reference patterns all of the members of which are co-original.

2. Apparatus for the classification of patterns as claimed in claim 1 in which said statistical property comprises at least one variance matrix of said plurality of reference representations.

3. Apparatus for the classification of patterns as claimed in claim 2 in which said reference patterns are handwritten numerals.

4. Apparatus for the classification of patterns as claimed in claim 2 in which said reference patterns are non-standard font numerals.

5. Apparatus for the classification of patterns as claimed in claim 2 in which at least some of said reference patterns are combinations of handwritten characters.

6. Apparatus for the classification of patterns as claimed in claim 2 in which at least some of said reference patterns are combinations of non-standard font characters.

7. Apparatus for the classification of patterns as claimed in claim 1 in which said statistical property is the signification variance matrix of said plurality of reference representations.

8. Apparatus for the classification of patterns as claimed in claim 1 in which said statistical property is the origination variance matrix of said plurality of reference representations.

9. Apparatus for the classification of patterns as claimed in claim 1 in which said statistical property is the overall variance matrix of said plurality of reference representations.

10. Apparatus for the classification of patterns as claimed in claim 1 in which said reference patterns are handwritten numerals.

11. Apparatus for the classification of patterns as claimed in claim 1 in which said reference patterns are non-standard font numerals.

12. Apparatus for the classification of patterns as claimed in claim 1 in which at least some of said reference patterns are combinations of handwritten characters.

13. Apparatus for the classification of patterns as claimed in claim 1 in which at least some of said reference patterns are combinations of non-standard font characters.