Robust image registration

• US 20070003165A1
• Filed: 06/19/2006
• Published: 01/04/2007
• Est. Priority Date: 06/20/2005
• Status: Active Grant

• Current Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Inventors: Bober, Miroslaw, Sibiryakov, Alexander
• Granted Patent: US 7,961,982 B2
• Time in Patent Office: Days
• US Class Current: 382/294
• CPC Class Codes: G06T 7/337 involving reference images or patches
• CPC Class Codes: G06T 7/35 using statistical methods
• CPC Class Codes: G06V 10/757 Matching configurations of points or features

Abstract

A method of estimating a transformation between a pair of images, comprises estimating local transformations for a plurality of regions of the images to derive a set of estimated transformations, and selecting a subset of said estimated local transformations as estimated global transformations for the image.

Claims

1. A method of estimating a transformation between a pair of images, comprising estimating local transformations for a plurality of regions of the images to derive a set of estimated transformations, and selecting a subset of said estimated local transformations as estimated global transformations for the image.

2. The method of claim 1 wherein the estimation of local transformations is capable of estimating motion additional to or other than translation, such as rotation and/or change of scale.

3. The method of claim 1 wherein said estimation of local transformations involves a parametric model having more than 2 parameters.

4. The method of claim 1 comprising further processing using the set of estimated transformations to derive at least one global transformation.

5. The method of claim 4 wherein the further processing comprises using a robust estimator such as RANSAC.

6. The method of claim 5 wherein the further processing further comprises least-squares estimation.

7. The method of claim 1 comprising using a similarity measure for comparing regions.

8. The method of claim 7 comprising using a similarity measure which is invariant to rotation and scale.

9. The method of claim 7 using the similarity measure for an estimation of transformation parameters.

10. The method of claim 7 wherein the similarity measure involves a correlation coefficient or phase correlation.

11. The method of claim 7 comprising, for each of a plurality of regions in the first image, comparing the region with each of a plurality of regions in the second image, to derive a plurality of similarity measures.

12. The method of claim 11 comprising, for each of the regions in the first image, selecting M regions, where M>

1, from the second image having the highest similarity measures.

13. The method of claim 12 using a voting scheme or accumulator for the plurality of local estimated transformations to select the subset.

14. The method of claim 13 comprising using the selected M regions in the voting scheme.

15. The method of claim 1 wherein the estimation of local transformations involves regions including feature or interest points, and matching of said regions and points.

16. The method of claim 15 wherein the estimation of local transformations involves, for each a plurality of points in the first image, matching the point and a surrounding local region with each of a plurality of points and a surrounding local image in the second image.

17. The method of claim 15 comprising using selected subset to select corresponding pairs of matched points in the first and second images.

18. The method of claim 15 comprising ordering the feature or interest points, each associated with a confidence measure, on the basis of confidence measure and location in the image.

19. The method of claim 1 comprising estimating the local transformations as rotation scale translation transformations.

20. The method of claim 1 comprising iterating the transformation estimations using search constraints, and modifying the search constraints.

21. The method of claim 1 comprising using a modified log polar transform of the image region, in which a log polar transform is applied to an area starting from a radial value r₀ where r₀>

0.

22. The method of claim 1, using a multiscale representation of the images.

38. The method of claim 36 further comprising the features of claim 2.

39. Apparatus for executing the method of claim 1.

40. A control device programmed to execute the method of claim 1.

41. A computer program, system or computer-readable storage medium for executing a method of claim 1.

23. A method of representing an image comprising ordering a plurality of feature or interest points, each associated with a confidence measure, on the basis of both confidence and location in the image.

24. The method of claim 23 wherein the points are ordered for the purposes of matching.

25. The method of claim 23 wherein the ordering on the basis of location is ordering on the basis of distance between points.

26. The method of claim 25 wherein the points are ordered on the basis of decreasing distance between points.

27. The method of claim 23 wherein the points are ordered on the basis of decreasing confidence.

28. The method of claim 23 wherein initial points are representative of the spatial distribution of the points throughout the image.

29. The method of claim 23 comprising dividing the image into regions, and ordering the points according to confidence and region.

30. The method of claim 29 comprising ordering the points according to confidence in each region followed by ordering by region.

31. The method of claim 23 wherein the confidence measure reflects the strength of the feature or interest point.

32. The method of claim 23 wherein the confidence measure reflects the probability of the feature or interest point matching a feature or interest point in another image.

35. A method of matching a pair of images using representations of the images derived in accordance with claim 23.

36. The method of claim 35 for estimating a transformation between a pair of images.

37. The method of claim 36 comprising estimating local transformations for a plurality of regions of the images to derive a set of estimated transformations, and selecting a subset of said estimated local transformations as estimated global transformations of the image.

33. A method of representing an image using a modified log polar transform in which a log polar transform is applied to an area starting from a radial value r₀, where r₀>

0.

34. The method of claim 33 wherein the modified log polar transform is such that the inverse transform is of the form of equations (12) and (13*).