SYSTEM AND METHOD FOR CHARACTERIZING HANDWRITTEN OR TYPED WORDS IN A DOCUMENT

• US 20090060335A1
• Filed: 08/30/2007
• Published: 03/05/2009
• Est. Priority Date: 08/30/2007
• Status: Active Grant

• Current Assignee: Majandro LLC (IP Edge LLC)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Perronnin, Florent C., Rodriguez Serrano, Jose A.
• Granted Patent: US 8,014,603 B2
• Time in Patent Office: Days
• US Class Current: 382/177
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/18086 by performing operations within image blocks or by using histograms
• CPC Class Codes: G06V 30/182 by coding the contour of the pattern
• CPC Class Codes: G06V 30/19073 Comparing statistics of pixel or of feature values, e.g. histogram matching
• CPC Class Codes: G06V 30/347 Sampling; Contour coding; Stroke extraction
• CPC Class Codes: G06V 40/388 Sampling; Contour coding; Stroke extraction

Abstract

A method of characterizing a word image includes traversing the word image stepwise with a window to provide a plurality of window images. For each of the plurality of window images, the method includes splitting the window image to provide a plurality of cells. A feature, such as a gradient direction histogram, is extracted from each of the plurality of cells. The word image can then be characterized based on the features extracted from the plurality of window images.

Claims

1. A method of characterizing a word image comprising:

traversing the word image stepwise with a window to provide a plurality of window images;

for each of the plurality of window images;

splitting the window image to provide a plurality of cells;

extracting a feature from each of the plurality of cells, the feature comprising a gradient orientation histogram; and

characterizing the word image based on the features extracted from the plurality of window images.

2. The method of claim 1, further comprising:

for each window image, determining a features vector based on the extracted features of each of the plurality of cells; and

characterizing the word image based on the features vectors of the plurality of window images.

3. The method of claim 1, wherein the splitting the window image comprises splitting at least a portion of the window image into an array of cells.

4. The method of claim 3, wherein the array comprises an M×

N array where N is a number of cells arranged in a horizontal direction and M is a number of cells arranged in a vertical direction.

5. The method of claim 4, wherein M and N are independently at least two.

6. The method of claim 3, wherein the portion of the window image that is split into cells bounds at least the active pixels in the window image.

7. The method of claim 1 wherein the cells are rectangular.

8. The method of claim 1, wherein the cells of a window image are of the same size.

9. The method of claim 1, wherein the extraction of a feature comprises computing a gradient direction histogram for the pixels in the cell.

10. The method of claim 9, wherein the computing of the gradient direction histogram comprises computing a gradient magnitude m and direction θ

for each pixel with coordinates (x,y) in the cell as;

$m<br><br>\left(x,y\right)=\sqrt{G_x^2+G_y^2}$ $\mathrm{\theta <br><br>}<br><br>\left(x,y\right)=a<br><br><br><br>\tan <br><br><br><br>2<br><br>\left(\frac{G_y}{G_x}\right),$

where atan2 is a function that gives the angle of the vector (G_x, G_y) in the range [−

π

,π

].

11. The method of claim 2, wherein the computing of the features vector comprises concatenating the extracted features.

12. The method of claim 11, wherein the computing of the features vector further comprises normalizing the features vector such that all of the component values sum to a fixed value.

13. The method of claim 1, wherein the word image comprises a bitmap acquired by segmenting a document image.

14. The method of claim 1, wherein the word image is characterized without characterization of individual characters of a character string within the word image.

15. A computer program product encoding instructions, which when executed on a computer causes the computer to perform the method of claim 1.

16. A processing system which executes instructions stored in memory for performing the method of claim 1.

17. A method of characterizing a document image comprising:

segmenting the document image to identify word images;

for an identified word image, traversing the word image stepwise with a window to provide a plurality of window images;

for each of the plurality of window images;

splitting the window image to provide a plurality of cells;

extracting a feature from each of the plurality of cells; and

computing a features vector based on the extracted features;

characterizing the word image based on the features vectors of the plurality of window images; and

characterizing the document based on the characterization of at least one of the identified word images.

18. The method of claim 17, further comprising classifying the characterized word image with a classifier trained to identify a keyword.

19. The method of claim 18, wherein the document image is an image of a handwritten document.

20. A processing system comprising:

a document segmentor which processes an input document image to identify word images;

a features extractor which extracts features of an identified word image and computes features vectors therefrom, the features extractor executing instructions for traversing the word image stepwise with a window to provide a plurality of window images and, for each of the plurality of window images, splitting the window image to provide a plurality of cells, extracting a feature from each of the plurality of cells, and computing a features vector based on the extracted features; and

a classifier which classifies the word image based on the computed features vectors of the window images.

21. The processing system of claim 20, wherein the classifier comprises a hidden Markov model trained to identify at least one keyword.