Local scale, rotation and position invariant word detection for optical character recognition

1. A computer-implemented method comprising:

receiving an input image;

generating progressively blurred images from the input image, by filtering the input image with different scales, until words in a blurred image appear as blobs;

detecting the blobs in the blurred image;

outputting ellipses over the blobs;

detecting, using one or more computing devices, a word in the input image based on the ellipses;

determining an orientation of the word;

normalizing, using the one or more computing devices, the orientation of the word;

performing optical character recognition on the word; and

outputting, using the one or more computing devices, text associated with the word.

2. The method of claim 1 , further comprising:

generating a first version of the word and an inverted version of the word;

generating a confidence score for each of the first version of the word and the inverted version of the word; and

wherein the text is associated with one of the first version or the inverted version of the word that has a higher confidence score.

3. The method of claim 1 , further comprising determining a representative ellipse for the word.

4. The method of claim 3 , wherein determining the representative ellipse for the word comprises:

removing ellipses that are non-words using word filters;

retaining at least one ellipse for the word in the input image; and

determining the representative ellipse for the word based on an aspect ratio of the at least one ellipse.

5. The method of claim 1 , further comprising:

comparing the text associated with the word to a dictionary definition; and

adjusting the text based on the comparison.

6. The method of claim 1 , wherein normalizing the orientation of the word comprises

normalizing the word to an upright orientation.

7. The method of claim 1 , further comprising outputting the text associated with the word for performing an image match and retrieving a document that includes the text.

8. The method of claim 1 , further comprising outputting the text associated with the word for performing translation on the text.

9. A system comprising:

one or more processors;

a controller stored on a memory and executable by the one or more processors, the controller configured to receive an input image;

a word detection engine configured to generate progressively blurred images from the input image, by filtering the input image with different scales, until words in a blurred image appear as blobs, to detect the blobs in the blurred image, to output ellipses over the blobs and to detect a word in the input image based on the ellipses;

a word orientation engine configured to determine an orientation of the word and to normalize the orientation of the word; and

an optical character recognition (OCR) engine configured to perform optical character recognition on the word and output text associated with the word.

10. The system of claim 9 , wherein the word orientation engine is further configured to:

generate a first version of the word and an inverted version of the word;

generate a confidence score for each of the first version of the word and the inverted version of the word; and

wherein the text is associated with one of the first version or the inverted version of the word that has a higher confidence score.

11. The system of claim 9 , wherein the word detection engine is further configured to determine a representative ellipse for the word.

12. The system of claim 11 , wherein determining the representative ellipse for the word comprises:

removing ellipses that are non-words using word filters;

retaining at least one ellipse for the word in the input image; and

determining the representative ellipse for the word based on an aspect ratio of the at least one ellipse.

13. The system of claim 9 , wherein the OCR engine is further configured to:

compare the text associated with the word to a dictionary definition; and

adjust the text based on the comparison.

14. The system of claim 9 , wherein normalizing the orientation of the word comprises

normalizing the word to an upright orientation.

15. A computer program product comprising a non-transitory computer useable medium including a computer readable program, wherein the computer readable program when executed on a computer causes the computer to:

receive an input image;

generate progressively blurred images from the input image, by filtering the input image with different scales, until words in a blurred image appear as blobs;

detect the blobs in the blurred image;

output ellipses over the blobs;

detect a word in the input image based on the ellipses;

determine an orientation of the word;

normalize the orientation of the word;

perform optical character recognition on the word; and

output text associated with the word.

16. The computer program product of claim 15 , further comprising:

generating a first version of the word and an inverted version of the word;

generating a confidence score for each of the first version of the word and the inverted version of the word; and

wherein the text is associated with one of the first version or the inverted version of the word that has a higher confidence score.

17. The computer program product of claim 16 , further comprising determining a representative ellipse for the word.

18. The computer program product of claim 17 , wherein determining the representative ellipse for the word comprises:

removing ellipses that are non-words using word filters;

retaining at least one ellipse for the word in the input image; and

determining the representative ellipse for the word based on an aspect ratio of the at least one ellipse.

19. The computer program product of claim 15 , wherein the computer readable program when executed on the computer further causes the computer to:

compare the text associated with the word to a dictionary definition; and

adjust the text based on the comparison.

20. The computer program product of claim 15 , wherein normalizing the orientation of the word comprises

normalizing the word to an upright orientation.