Image retrieval

• US 20040202385A1
• Filed: 04/09/2003
• Published: 10/14/2004
• Est. Priority Date: 04/09/2003
• Status: Active Grant

• Current Assignee: ArcSoft, Inc.
• Original Assignee: ArcSoft, Inc.
• Inventors: Cheng, Min, Xiong, Wei, Zhou, Lingxiang
• Granted Patent: US 7,519,236 B2
• Time in Patent Office: Days
• US Class Current: 382/305
• CPC Class Codes: G06V 10/32 Normalisation of the pattern dimensions

Abstract

A method for searching an image database includes capturing an image of a photograph and a background, determining a boundary of the photograph in the image, cropping the photograph from the image, correcting the perspective of the photograph, compensating colors of the photograph, and matching the photograph with an image in the image database.

Claims

1. A method for searching an image database, comprising:

capturing a first image of a photograph and a background using an image input device;

determining a boundary of the photograph from the background in the first image;

cropping the photograph from the first image; and

matching the photograph with a second image in the image database.

2. The method of claim 1, wherein said determining a boundary of the photograph in the first image comprises:

capturing a third image of the background without the photography using the image input device;

determining differences between the first image and the third image, the differences defining a region;

determining a perimeter of the region; and

determining boundary lines from the perimeter, the boundary lines defining the boundary of the photograph.

3. The method of claim 2, wherein said determining a perimeter of the region comprises:

scanning each of horizontal and vertical lines that comprise the region; and

reserving a first pixel and a last pixel in each of the horizontal and vertical lines as pixels that comprise the perimeter of the region.

4. The method of claim 2, wherein said determining boundary lines comprises performing Hough Transformation to detect the boundary lines from the perimeter.

5. The method of claim 4, wherein said determining the boundary lines further comprises:

invalidating a boundary line if the boundary line is too close to a corresponding edge of the first image, the boundary line is substantially parallel to the corresponding edge, and a confidence in the boundary line'"'"'s subtense is high enough; and

searching for another boundary line.

6. The method of claim 5, wherein said searching for another boundary line comprises:

using Hough Transformation to detect said another boundary line, wherein the angle of said another boundary line is limited within a range of the angle of the invalid boundary line'"'"'s subtense.

7. The method of claim 1, further comprising rotating the photograph after said cropping and before said matching.

8. The method of claim 7, wherein said rotating the photograph comprises performing an Affine Transformation.

9. The method of claim 7, further comprising, prior to said matching the photograph, resizing the photograph.

10. The method of claim 1, wherein said matching the photograph comprises using at least one of edge feature matching, texture feature matching, and color feature matching to match the photograph to the images in the database.

11. The method of claim 1, further comprising compensating colors of the photograph before said matching.

12. The method of claim 11, wherein said compensating colors of the photograph comprises performing histogram equalization to the photograph.

13. The method of claim 11, wherein said compensating colors of the photograph comprises performing color constancy to the photograph.

14. The method of claim 13, wherein said performing color constancy comprises:

transforming the photograph using an inverse homography matrix to approximate RGB values of the photography under a first lighting condition.

15. The method of claim 14, wherein said performing color constancy further comprises:

generating a physical copy of a reference image, wherein the reference image is assumed to be captured under the first lighting condition;

capturing a test image of the physical copy under a second lighting condition with the image input device;

determining a homography matrix that transforms RGB values of the reference image under the first lighting condition to RGB values of the physical copy under the second lighting condition; and

inverting the homography matrix to generate the inverse homography matrix.

16. The method of claim 15, wherein the reference image comprises a plurality of blocks having different colors.

17. The method of claim 1, wherein the image input device is selected from the group consisting a digital video camera, a digital still camera, and a scanner.

18. The method of claim 9, further comprising, transmitting the first image of from a client computer over a network to a server computer.

19. The method of claim 18, wherein the server computer performs said determining a boundary of the photograph, said cropping the photograph, said rotating the photography, said resizing the photography, and said matching the photograph.

20. The method of claim 18, further comprising, transmitting the second image from the server computer to the client computer.