OCR-GUIDED TEXT TOKENIZATION OF DIGITAL IMAGES

• US 20100150460A1
• Filed: 12/16/2008
• Published: 06/17/2010
• Est. Priority Date: 12/16/2008
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Fan, Zhigang, Bai, Yingjun, Tse, Francis, Campanelli, Michael R.
• Granted Patent: US 8,180,164 B2
• Time in Patent Office: Days
• US Class Current: 382/232
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/262 using context analysis, e.g. lexical, syntactic or semantic context

Abstract

An image processing method in which OCR is used to guide the text tokenization. More particularly, OCR is first performed on each symbol in the scanned image. For example, a symbol may be a number, letter, or other character. During the tokenization process, the OCR results are used to select appropriate matching criteria for each symbol. The symbols that are recognized as different characters are not allowed to be clustered into the same group. The symbols with the same OCR results are clustered according to the recognition confidence levels.

Claims

1. An image processing method comprising:

scanning an input image via an image input device;

compressing the scanned image using an image compression tool by performing OCR (Optical Character Recognition) on each symbol in the scanned image to generate OCR results and then performing tokenization on the scanned image using the OCR results; and

storing the compressed image in a storage device or printing the compressed image via an image output device after it has been decoded.

2. The method defined in claim 1, wherein the image compression tool comprises JBIG2.

3. The method defined in claim 1, wherein during the tokenization process, the OCR results are used to select appropriate matching criteria for each symbol, the symbols that are recognized as different characters are not allowed to be clustered into the same group, and the symbols with the same OCR results are clustered according to a pre-defined set of recognition confidence levels.

4. The method defined in claim 2, wherein the following rules are applied during the tokenization process:

the symbols with different primary OCR results are not clustered into the same group; and

for symbols with the same primary OCR result;

if both symbols have high confidence levels for primary results and low confidence levels for secondary results, use a loose matching criteria to allow large shape variation;

if both symbols have high confidence levels for primary results and at least one has a high confidence levels for its secondary results, use a tight matching criteria to avoid misclassification;

if at least one symbol has a low confidence level for its primary result, use a moderate to tight matching criteria.

5. The method defined in claim 1, further comprising:

performing pre-filtering on the scanned image to remove noise in the background or to clean up noise on edges; and

guiding the level of pre-filtering on a page basis using OCR results, wherein the level of pre-filtering can be set for a whole page or selected areas with the help of image segmentation.

6. The method defined in claim 1, further comprising:

finding a text string with OCR; and

comparing the text string before and after any lossy image processing operations such as pre-filtering or JBIG2 compression to determine whether the same text string is generated by the OCR process.

7. An image processing system comprising:

an image input device for scanning an input image;

an image processing device connected to the image input device and operative to compress the scanned image using an image compression tool by performing OCR (Optical Character Recognition) on the scanned image to generate OCR results and then performing tokenization on the scanned image using the OCR results;

a storage device connected to the image processing device and operative to store the compressed image; and

an image output device connected to the image processing device and operative to print the compressed image after it has been decoded.

8. The system defined in claim 7, wherein the image compression tool comprises JBIG2.

9. The system defined in claim 7, wherein the image processing device is further operative to use the OCR results to select appropriate matching criteria for each symbol, not allow the symbols that are recognized as different characters to be clustered into the same group, and cluster the symbols with the same OCR results according to a pre-defined set of recognition confidence levels.

10. The system defined in claim 7, wherein the following rules are applied during the tokenization process:

the symbols with different primary OCR results are not clustered into the same group; and

for symbols with the same primary OCR result;

if both symbols have high confidence levels for primary results and low confidence levels for secondary results, use a loose matching criteria to allow large shape variation;

if both symbols have high confidence levels for primary results and at least one has a high confidence levels for its secondary results, use a tight matching criteria to avoid misclassification;

if at least one symbol has a low confidence level for its primary result, use a moderate to tight matching criteria.

11. The system defined in claim 7, wherein the image processing device is further operative to:

perform pre-filtering on the scanned image to remove noise in the background or to clean up noise on edges; and

guide the level of pre-filtering on a page basis using OCR results, wherein the level of pre-filtering can be set for a whole page or selected areas with the help of image segmentation.

12. The system defined in claim 7, wherein the image processing device is further operative to:

find a text string with OCR; and

compare the text string before and after any lossy image processing operations such as pre-filtering or JBIG2 compression to determine whether the same text string is generated by the OCR process.

13. A computer program product comprising:

a computer-usable data carrier storing instructions that, when executed by a computer, cause the computer to perform a method comprising;

scanning an input image via an image input device;

compressing the scanned image using an image compression tool by performing OCR (Optical Character Recognition) on the scanned image to generate OCR results and then performing tokenization on the scanned image using the OCR results; and

storing the compressed image in a storage device or printing the compressed image via an image output device after it has been decoded.

14. The computer program product defined in claim 13, wherein the image compression tool comprises JBIG2.

15. The computer program product defined in claim 13, wherein during the tokenization process, the OCR results are used to select appropriate matching criteria for each symbol, the symbols that are recognized as different characters are not allowed to be clustered into the same group, and the symbols with the same OCR results are clustered according to a pre-defined set of recognition confidence levels.

16. The computer program product defined in claim 13, wherein the following rules are applied during the tokenization process:

the symbols with different primary OCR results are not clustered into the same group; and

for symbols with the same primary OCR result;

if both symbols have high confidence levels for primary results and low confidence levels for secondary results, use a loose matching criteria to allow large shape variation;

if both symbols have high confidence levels for primary results and at least one has a high confidence levels for its secondary results, use a tight matching criteria to avoid misclassification;

if at least one symbol has a low confidence level for its primary result, use a moderate to tight matching criteria.

17. The computer program product defined in claim 13, wherein the method further comprises:

performing pre-filtering on the scanned image to remove noise in the background or to clean up noise on edges; and

guiding the level of pre-filtering on a page basis, wherein the level of pre-filtering can be set for a whole page or selected areas with the help of image segmentation.

18. The computer program product defined in claim 13, wherein the method further comprises:

finding a text string with OCR; and

comparing the text string before and after any lossy image processing operations such as pre-filtering or JBIG2 compression to determine whether the same text string is generated by the OCR process.