Apparatus and method for improving recognition of patterns by prototype transformation

• US 5,422,961 A
• Filed: 04/03/1992
• Issued: 06/06/1995
• Est. Priority Date: 04/03/1992
• Status: Expired due to Term

• Current Assignee: AT&T IPM Corp. (AT&T, Inc.)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Simard, Patrice Y.
• Primary Examiner(s): Mancuso, Joseph
• Application Number:
  US07/862,696
• Time in Patent Office: 1,159 Days
• Field of Search: 382/27, 382/36, 382/14, 382/15, 382/41, 382/37, 382/38, 382/39, 395/3, 395/61
• US Class Current: 382/224
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 10/76 based on eigen-space representations, e.g. from pose or different illumination conditions; Shape manifolds

Abstract

Speed and accuracy of the recognition process for an alphanumeric automated recognizer is enhanced by a novel scheme for comparing known prototype symbols to examples of unknown symbol. The scheme is invariant with respect to a selected set of small transformations of either the prototypes or the examples, requires only a few operators, and generates accuracy rates comparable to a human operator. The small transformations of interest are expressed locally by a process in which, during the data processing, the derivative of the transformed image with respect to the parameter that controls the transformation is calculated. This directional derivative then is used as the generator of interest in the recognition machine and is incorporated into the comparison process. The derivatives are generated by using tangent plane concepts, which yield a useful approximation of an otherwise not quantifiable complex surface in the neighborhood occupied by the unknown example. Computation time is decreased and recognition accuracy is increased.

Claims

1. A pattern-classifying machine for assigning identification labels to unknown input patterns such that each input pattern is classified as a member of only one of several labeled prototype classes, said classifying machine comprising (a) a database consisting of data files of patterns which are known members of each prototype class;

(b) a pattern comparator;

(c) means for inputting said input patterns and said known members to said comparator; and

(d) means for making predetermined transformations of prototype patterns and of input patterns, the improvement comprising;

means for generating tangent vector patterns directly from selected transformations of each said prototype pattern, wherein;

each selected transformation leaves the prototype label invariant;

each selected transformation operates on image vectors in a vector space that represents pixel values; and

each of the tangent vector patterns is represented in the vector space as a vector proportional to a difference between transformed and original patterns;

means for generating for each said prototype pattern a transformed prototype pattern, by adding to each said prototype pattern selected linear combinations of said first tangent vector patterns;

means for computing the Euclidean distance between each of said transformed prototype patterns and a one of said;

input patterns and

means for determining the smallest of the just-computed Euclidean distances, the smallest said distance determining which labeled prototype pattern is closest to said input pattern.

3. Apparatus in accordance with claims 1 or 2, wherein said transformed patterns are formed by rotation.

4. Apparatus in accordance with claims 1 or 2, wherein said transformed patterns are formed by skewing.

5. Apparatus in accordance with claims 1 or 2, wherein said transformed patterns are formed by translation.

6. Apparatus in accordance with claims 1 or 2, wherein said transformed patterns are formed by line thickness change.

7. Apparatus in accordance with claims 1 or 2, wherein said transformed patterns are formed by grey-level changes.

8. Apparatus in accordance with claims 1 or 2, wherein said transformed patterns are formed by alphanumeric symbols.

9. Apparatus in accordance with claim 8, wherein said patterns comprise alphanumeric address information on postal items to be sorted for ultimate destination.

10. Apparatus in accordance with claim 9, further comprising a video scanner, means for transporting said address information past said scanner, means for connecting said scanned information into individual pixel maps of each scanned unknown alphanumeric symbol, and a memory store fore receiving determinations of the labeled prototype which is closest as a pattern to each said unknown example.

2. A pattern-classifying machine for assigning identification labels to unknown input patterns such that each input pattern is classified as being a member of only one of several labeled prototype classes, said classifying machine comprising (a) a database consisting of data files of patterns which are known members of each prototype class;

(b) a pattern comparator;

(c) means for inputting said input patterns and said known members to said comparator; and

(d) means for making predetermined transformations of prototype patterns and of input patterns, the improvement comprising;

means for generating tangent vector patterns directly from selected transformations of each said input pattern, wherein;

each selected transformation leaves the input pattern invariant as to its class;

each selected transformation operates on image vectors in a vector space that represents pixel values; and

each of the tangent vector patterns is represented in the vector space as a vector proportional to a difference between transformed and original patterns;

means for generating for each said input pattern a transformed input pattern, by adding to each said input pattern selected linear combinations of said tangent vector patterns;

means for computing the Euclidean distance between each of said transformed example patterns and a one of said prototype patterns; and

means for determining the smallest of the just-computed Euclidean distances, the smallest said distance determining which labeled prototype pattern is closest to said input pattern.