System and method for character recognition with normalization

• US 5,633,954 A
• Filed: 09/30/1994
• Issued: 05/27/1997
• Est. Priority Date: 05/18/1993
• Status: Expired due to Term

• Current Assignee: Massachusetts Institute of Technology
• Original Assignee: Massachusetts Institute of Technology
• Inventors: Gupta, Amar, Wang, Patrick S. P., Nagendraprasad, Maram V.
• Primary Examiner(s): Mancuso, Joseph
• Assistant Examiner(s): DEL ROSSO, GERARD DOMNICK
• Application Number:
  US08/316,718
• Time in Patent Office: 970 Days
• Field of Search: 382/181, 382/187, 382/276, 382/296, 382/289, 382/298, 382/290, 382/302, 382/258, 382/259, 382/155, 382/156
• US Class Current: 382/187
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1478 of characters or characters lines
• CPC Class Codes: G06V 30/166 Normalisation of pattern dimensions
• CPC Class Codes: G06V 30/194 References adjustable by an adaptive method, e.g. learning

Abstract

A character recognition system comprises means for receiving a first binary image of a plurality of characters; segmenting means for separating the first binary image into a plurality of second binary images corresponding to the plurality of characters in the first binary image; normalizing means for reducing variations among the characters, which normalizing means comprises stroke adjusting means for processing the second binary images to adjust strokes of the characters to a predetermined thickness, size normalization means for processing the second binary images to proportionally adjust the overall size of the characters to a predetermined overall size, and slant correction means for processing the second binary images to reduce a slant with respect to a predetermined axis of the characters; and recognizer means coupled to the normalizing means for processing the normalized second binary images to determine which character is represented by each of the normalized second binary images.

Claims

1. A neural network-based character recognition system, comprising:

a receiver for receiving a first binary image of a plurality of characters;

a segmenter for separating the first binary image into a plurality of second binary images corresponding to the plurality of characters in the first binary image;

a normalizer for reducing variations among the plurality of characters by processing the second binary images, comprising;

a stroke adjuser for processing the second binary images to adjust strokes of the characters first to a one-pixel thickness and then to adjust the one-pixel thickness strokes to a second thickness which is substantially uniform and which is greater than the one-pixel thickness,

a size normalizer for processing the second binary images to proportionally adjust the overall size of the characters to a predetermined overall size, and

a slant correcter for processing the second binary images to reduce a slant with respect to a predetermined axis of the characters by using a binary search strategy which converges within a predetermined error margin; and

a neural network recognizer coupled to the normalizer for processing the stroke-adjusted, size-normalized, and slant corrected second binary images to determine which character is represented by each of the second binary images processed by the normalizer.

2. The character recognition system of claim 1 further comprising a scanner for scanning a document having the plurality of characters thereon to create the first binary image.

3. The character recognition system of claim 2 wherein the document comprises a financial document.

4. The character recognition system of claim 3 wherein the financial document comprises a check.

5. The character recognition system of claim 1 wherein the plurality of characters comprises generally non-cursive, handwritten characters.

6. The character recognition system of claim 5 wherein the plurality of characters includes Arabic numerals.

7. The character recognition system of claim 1 wherein the neural network identifies one character from a predetermined set of characters for each of the second binary images processed by the normalizer.

8. The character recognition system of claim 7 wherein the neural network generates a confidence value for each of the characters identified from the predetermined set of characters to indicate a measure of the accuracy of the identification made by the neural network.

9. The character recognition system of claim 8 wherein the confidence value ranges from zero to one, a value of zero indicative of the worst possible accuracy of the identification and a value of one indicative of the best possible accuracy.

10. The character recognition system of claim 1 wherein the segmenter comprises a plurality of stages.

11. A method of processing a binary image of a character and then providing the processed binary image to a neural network character recognizer for recognition by the neural network character recognizer as one of a predetermined set of characters, comprising the steps of:

providing the binary image of the character which includes one or more strokes forming the character, each stroke having a length and a thickness which can vary over the length;

processing the binary image to adjust each of the strokes of the character to a one-pixel thickness;

processing the adjusted binary image to adjust each of the adjusted one-pixel thickness strokes of the character to a second thickness which is substantially uniform and which is greater than the one-pixel thickness; and

providing the twice-adjusted binary image to the neural network character recognizer for recognition as one of the predetermined set of characters.