Unconstrained handwriting recognition

• US 7,164,794 B2
• Filed: 08/22/2002
• Issued: 01/16/2007
• Est. Priority Date: 08/22/2002
• Status: Active Grant

• Current Assignee: Winbond Electronics Corporation
• Original Assignee: Winbond Electronics Corporation
• Inventors: Li, Guoxing, Shi, Bingxue
• Primary Examiner(s): DESIRE, GREGORY M
• Application Number:
  US10/225,198
• Publication Number:
  US 20040037464A1
• Time in Patent Office: 1,608 Days
• Field of Search: 382/112, 382/156, 382/190, 382/224, 382/232, 382/159, 706/16, 706/20
• US Class Current: 382/190
• CPC Class Codes: G06V 10/955 using specific electronic processors
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/18076 by analysing connectivity, e.g. edge linking, connected component analysis or slices

Abstract

Methods and systems of the present invention may be used to recognize digital image data arranged in rows and columns. Exemplary embodiments may include a feature extractor for extracting feature information from data representing the rows and columns of the digital image data, a feature compressor for compressing the extracted feature information, and a neural network for classifying the digital image data from the compressed, extracted feature information.

Claims

1. A system for recognizing digital image data arranged in rows and columns, comprising:

a feature extractor for extracting feature information from data representing the rows and columns of the digital image data, wherein the extracted feature information of a row comprises a plurality of row bits and the extracted feature information of a column comprises a plurality of column bits;

a feature compressor for compressing the extracted feature information, whereby a portion of the row bits from the extracted feature information of the rows is removed and a portion of the column bits from the extracted feature information of the columns is removed; and

a neural network for classifying the digital image data from the compressed, extracted feature information, wherein the feature extractor includes a shift register to extract feature information of a diagonal of the digital image data.

3. The system of claim 1 or 2, wherein the neural network includes a multi-layer perceptron comprising a sensory layer, association layer, and response layer, and wherein the sensory layer receives the compressed feature information.

4. The system of claim 3, wherein the compressed feature information is received in pipeline mode.

5. The system of claim 3, wherein the sensory layer comprises two groups of sensory neurons, the association layer comprises two groups of association neurons, the first group of sensory neurons is connected with the second group of association neurons, and the second group of sensory neurons is connected with the second group of association neurons.

6. The system of claim 5, wherein the response layer comprises response neurons and the two groups of association neurons are connected to the response neurons.

7. The system of claim 3, wherein the multi-layer perceptron is trained using a back propagation learning algorithm.

8. The system of claim 1 or 2, wherein the feature extractor includes connected component detection array to extract the feature information of each row and each column of the digital image data.

10. The system of claim 1 or 9, wherein the feature compressor combines the extracted feature information of the diagonal and each row, sends the combined feature information of the diagonal and each row to the feature extractor to extract additional row feature information, combines the extracted feature information of the diagonal and each column, and sends the combined feature information of the diagonal and each column to the feature extractor to extract additional column feature information.

11. The system of claim 10, wherein the feature compressor compresses the additional row feature information and the additional column feature information to produce the compressed, extracted feature information.

2. A system for recognizing digital image data arranged in rows and columns, comprising:

a feature extractor for extracting feature information from data representing the rows and columns of the digital image data, wherein the extracted feature information of a row comprises a plurality of row bits and the extracted feature information of a column comprises a plurality of column bits;

a feature compressor for compressing the extracted feature information, whereby a portion of the row bits from the extracted feature information of the rows is removed and a portion of the column bits from the extracted feature information of the columns is removed; and

a neural network for classifying the digital image data from the compressed, extracted feature information.

wherein the feature compressor logically ORs remaining row bits between neighboring rows and logically ORs remaining column bits between neighboring columns.

9. The system of claim 2, wherein the feature extractor extracts feature information reflecting at least one diagonal of the digital image data.