Methods and Systems for Identifying Text in Digital Images

• US 20080056573A1
• Filed: 09/06/2006
• Published: 03/06/2008
• Est. Priority Date: 09/06/2006
• Status: Active Grant

• Current Assignee: Sharp Kabushiki Kaisha (Hon Hai Precision Industry Co., Ltd.)
• Original Assignee: Sharp Kabushiki Kaisha (Hon Hai Precision Industry Co., Ltd.)
• Inventors: Matsuda, Toyohisa, Campbell, Richard John, Chen, Lawrence Shao-hsien
• Granted Patent: US 7,876,959 B2
• Time in Patent Office: Days
• US Class Current: 382/176
• CPC Class Codes: G06V 30/413 Classification of content, e.g. text, photographs or tables

Abstract

Embodiments of the present invention comprise systems, methods and devices for detection of text in an image using an initial text classification result and a verification process.

Claims

1. A method for identifying text in a digital image, said method comprising:

a) receiving a first feature data set comprising measurements of a first feature on a digital image;

b) receiving a confidence data set related to said first feature data set, wherein said confidence data set indicates a confidence level of the measurements of said first feature data set;

c) receiving digital image data related to said digital image;

d) calculating a pictorial feature for said digital image data using said first feature data set and said confidence data set;

e) forming a pictorial confidence image using said pictorial feature and said first feature data set; and

f) generating a text-map based on said pictorial confidence image.

2. A method as described in claim 1 wherein said pictorial feature is masked entropy.

3. A method as described in claim 1 wherein said first feature is an edge feature.

4. A method as described in claim 1 wherein said digital image data is a luminance image.

5. A method as described in claim 1 wherein the resolution of said digital image data is less than the resolution of said first feature data set.

6. A method as described in claim 1 further comprising generating a mask from said first feature data set and said confidence data set.

7. A method as described in claim 1 wherein said forming a pictorial confidence image further comprises hysteresis thresholding.

8. A method as described in claim 1 wherein said forming a pictorial confidence image further comprises a plurality of scan passes.

9. A method as described in claim 8 wherein said number of scan passes is four.

10. A method as described in claim 1 wherein said pictorial confidence image and said first feature data set have the same resolution.

11. A method for identifying text in a digital image, said method comprising:

a) expanding the support region of a candidate text pixel in a text-candidate map, thereby producing a revised text-candidate map;

b) discriminating pictorial regions in said digital image based on an entropy measure comprising masking using said revised text-candidate map; and

c) refining said revised text-candidate map based on said pictorial regions.

12. A method as described in claim 11 wherein said discriminating further comprises averaging said entropy measure.

13. A method as described in claim 11 wherein said discriminating further comprises hysteresis thresholding.

14. A method as described in claim 11 wherein said expanding comprises:

a) receiving an edge map wherein said edge map identifies edges in said digital image;

b) generating an edge count wherein said edge count generating comprises associating an entry in said edge count with a neighborhood in said edge map and the value of said entry is the sum of the edge pixels in said neighborhood;

c) receiving a text-candidate map wherein said text-candidate map identifies text-candidate pixels in said digital image; and

d) generating a text count wherein said text count generating comprises associating an entry in said text count with a neighborhood in said text-candidate map and the value of said entry is the sum of the text-candidate pixels in said neighborhood.

15. A method as described in claim 14 wherein said candidate-text map is based on a texture feature of said digital image.

16. A system for identifying text in a digital image, said system comprising:

a) an expander for expanding the support region of a candidate text pixel in a text-candidate map, thereby producing a revised text-candidate map;

b) a discriminator for discriminating pictorial regions in said digital image based on an entropy measure comprising masking using said revised text-candidate map; and

c) a refiner for refining said revised text-candidate map based on said pictorial regions.

17. A system as described in claim 16 wherein said discriminating further comprises averaging said entropy measure.

18. A system as described in claim 16 wherein said discriminating further comprises hysteresis thresholding.

19. A system as described in claim 16 wherein said expander comprises:

a) an edge map receiver for receiving an edge map wherein said edge map identifies edges in said digital image;

b) an edge count generator for generating an edge count wherein said edge count generating comprises associating an entry in said edge count with a neighborhood in said edge map and the value of said entry is the sum of the edge pixels in said neighborhood;

c) a text-candidate map receiver for receiving a text-candidate map wherein said text-candidate map identifies text-candidate pixels in said digital image; and

d) a text count generator for generating a text count wherein said text count generating comprises associating an entry in said text count with a neighborhood in said text-candidate map and the value of said entry is the sum of the text-candidate pixels in said neighborhood.

20. A system as described in claim 19 wherein said candidate-text map is based on a texture feature of said digital image.

21. A system as described in claim 16 wherein said discriminating further comprises a plurality of scan passes.