Deblurring of an image from a sequence of images

1. A method for deblurring an image from a sequence of images, the image comprising pixels, the method comprising:

determining motion estimates of the pixels of the image;

modeling point spread functions for the pixels of the image using the determined motion estimates;

determining a deblurred estimate of the image using the modeled point spread functions; and

iteratively improving the determined motion estimates for the pixels by reducing a pixel's blur model magnitude by one unit in each direction until the modeled point spread functions converge.

2. The method according to claim 1 , wherein determining motion estimates of the pixels of the image further comprises:

applying a block matching algorithm to the image and temporally neighboring images from the sequence of images; and

removing non-parallel motion estimates with a temporal coherence filter, said removing comprising comparing a dot product of a pixel's motion estimates relative to a temporally preceding image and a temporally successive image with Euclidean norms of those motion estimates.

3. The method according to claim 2 , wherein an output of the temporal coherence filter is ‘0’ if the dot product of a pixel's determined motion estimates is smaller than a sum of the squared Euclidean norms of those motion estimates multiplied by a threshold, and wherein the output is half the sum of those motion estimates otherwise.

4. The method according to claim 1 , wherein determining motion estimates of the pixels of the image comprises eliminating noise from the motion estimates with a median filter.

5. The method according to claim 1 , wherein determining motion estimates of the pixels of the image comprises smoothing an output motion field with a low-pass filter.

6. The method according to claim 1 , wherein improving the determined motion estimates for the pixels comprises back-projecting the deblurred estimate of the image.

7. The method according to claim 1 , wherein modeling point spread functions for the pixels of the image considers motion blur using a symmetric linear motion blur kernel and lens blur using a disk kernel.

8. An apparatus for deblurring an image from a sequence of images, the image comprising pixels, wherein the apparatus comprises:

a motion estimation stage configured to determine motion estimates of the pixels of the image;

a point spread function modeling stage configured to model point spread functions for the pixels of the image using the determined motion estimates;

a deconvolution stage configured to determine a deblurred estimate of the image using the modeled point spread functions; and

a feedback loop configured to iteratively improve the determined motion estimates for the pixels by reducing a pixel's blur model magnitude by one unit in each direction until the modeled point spread functions converge.

9. The apparatus according to claim 8 , wherein the motion estimation stage is configured to:

apply a block matching algorithm to the image and temporally neighboring images from the sequence of images; and

remove non-parallel motion estimates with a temporal coherence filter, which compares a dot product of a pixel's determined motion estimates relative to a temporally preceding image and a temporally successive image with Euclidean norms of those determined motion estimates.

10. The apparatus according to claim 9 , wherein an output of the temporal coherence filter is ‘0’ if the dot product of a determined pixel's motion estimates is smaller than a sum of the squared Euclidean norms of those motion estimates multiplied by a threshold, and wherein the output is half the sum of those determined motion estimates otherwise.

11. The apparatus according to claim 8 , wherein the motion estimation stage is configured to eliminate noise from the determined motion estimates with a median filter.

12. The apparatus according to claim 8 , wherein the motion estimation stage is configured to smooth an output motion field with a low-pass filter.

13. The apparatus according to claim 8 , wherein the feedback loop is configured to back-project the deblurred estimate of the image.

14. The apparatus according to claim 8 , wherein the point spread function modeling stage is configured to consider motion blur using a symmetric linear motion blur kernel and lens blur using a disk kernel.

15. A non-transitory computer readable storage medium having stored therein instructions enabling deblurring of an image from a sequence of images, the image comprising pixels, which when executed by a computer, cause the computer to:

determine motion estimates of the pixels of the image; model point spread functions for the pixels of the image using the motion estimates;

determine a deblurred estimate of the image using the modeled point spread functions; and

iteratively improve the determined motion estimates for the pixels by reducing a pixel's blur model magnitude by one unit in each direction until the modeled point spread functions converge.

16. The non-transitory computer readable storage medium according to claim 15 , wherein determining motion estimates of the pixels of the image comprises:

applying a block matching algorithm to the image and temporally neighboring images from the sequence of images; and

removing non-parallel motion estimates with a temporal coherence filter, which compares a dot product of a pixel's determined motion estimates relative to a temporally preceding image and a temporally successive image with Euclidean norms of those determined motion estimates.

17. The non-transitory computer readable storage medium according to claim 16 , wherein an output of the temporal coherence filter is ‘0’ if the dot product of a pixel's determined motion estimates is smaller than a sum of the squared Euclidean norms of those motion estimates multiplied by a threshold, and wherein the output is half the sum of those determined motion estimates otherwise.

18. The non-transitory computer readable storage medium according to claim 15 , wherein determining motion estimates of the pixels of the image comprises eliminating noise from the determined motion estimates with a median filter.

19. The non-transitory computer readable storage medium according to claim 15 , wherein determining motion estimates of the pixels of the image comprises smoothing an output motion field with a low-pass filter.

20. The non-transitory computer readable storage medium according to claim 15 , wherein improving the determined motion estimates for the pixels comprises back-projecting the deblurred estimate of the image.

21. The non-transitory computer readable storage medium according to claim 15 , wherein modeling point spread functions for the pixels of the image considers motion blur using a symmetric linear motion blur kernel and lens blur using a disk kernel.