Outlier detection for colour mapping

1. A method for outlier detection for color mapping, the method comprising:

retrieving a reference image and a test image;

determining a set of pairs of corresponding feature points between the reference image and the test image, for each pair of the determined set there is a feature point in the reference image and a corresponding feature point in the test image;

determining an initial color mapping model for each pair of corresponding feature points between the test image and the reference image;

applying the initial color mapping model to the test image to generate an initial color-corrected test image;

for each pair of corresponding feature points of the determined set of pairs, comparing color values of a spatial area around the feature point in the reference image with color values of a spatial area around the corresponding feature point the initial color-corrected test image;

when said comparison of color values related to a given pair of corresponding feature points meets a threshold criteria, deleting the given pair of corresponding feature points to generate a reduced set of pairs of corresponding feature points; and

processing the reduced set of corresponding feature points for transmission to a signal processor.

2. The method according to claim 1 , wherein the corresponding feature points are determined using Geometric Feature Correspondences.

3. The method according to claim 1 , wherein the reference image and the test image are images of a stereo pair.

4. The method according to claim 1 , further comprising:

determining a refined color mapping model from image colors of the reduced set of pairs of corresponding feature points for mapping image colors of the reduced set of pairs of corresponding feature points between the test image and the reference image;

applying the refined color mapping model to the test image to generate a refined color-corrected test image;

for each non-deleted pair of corresponding feature points, comparing color values of a spatial area around a feature point of a non-deleted pair, which belongs to the reference image, with color values of a corresponding spatial area around the other feature point of the non-deleted pair, which belongs to the refined color-corrected test image; and

when the comparing of color values meets a threshold criteria for a given non-deleted pair of corresponding feature points, deleting the given non-deleted pair of corresponding feature points to generate a further reduced set of pairs of corresponding feature points.

5. The method according to claim 1 , wherein determining the initial color mapping model comprises determining Color Correspondences.

6. The method according to claim 1 , wherein the threshold is determined as a variance of an estimation error determined as a difference between a reference color and an estimated color.

7. The method according to claim 1 , wherein said comparison of color values is based on an absolute difference between mean color values of colors in said corresponding spatial areas for each color channel, and wherein deleting a pair is performed if said absolute difference exceeds a threshold for a majority of color channels.

8. The method according to claim 1 , wherein the initial color mapping model is a color transform.

9. The method according to claim 1 , wherein the initial color mapping model is a gamma-offset-gain-model.

10. The method according to claim 1 , wherein said comparison of color values is performed directly pixel to pixel in said corresponding spatial areas to decide whether pixels of each pixel to pixel comparison match together, and wherein deleting a pair is performed if a total number of pixel to pixel matches in the corresponding spatial areas of the given pair of corresponding feature points is below a threshold.

11. The method according to claim 1 , comprising clustering colors of corresponding spatial areas of each pair, wherein said comparison of color values is based on a difference computed between colors of color clusters that correspond in said corresponding spatial areas and wherein deleting a pair is performed if the difference computed for the given pair of corresponding feature points exceeds a threshold.

12. An image processing apparatus for outlier detection for color mapping, the image processing apparatus comprising:

a memory; and

at least one processor configured to:

determine a set of pairs of corresponding feature points between a reference image and a test image, for each pair of the determined set there is a feature point in the reference image and a corresponding feature point in the test image,

determine an initial color mapping model for each pair of corresponding feature points between the test image and the reference image,

apply the initial color mapping model to the test image to generate an initial color-corrected test image;

for each pair of corresponding feature points of the determined set of pairs, compare color values of a spatial area around the feature point in the reference image with color values of a spatial area around the corresponding feature point of the initial color-corrected test image;

delete a given pair of corresponding feature points to generate a reduced set of pairs of corresponding feature points, when said comparison of color values related to the given pair of corresponding feature points meets a threshold criteria; and

process the reduced set of pairs of corresponding feature points for transmission to a signal processor.

13. The image processing apparatus according to claim 12 , wherein the at least one processor is configured to determine the corresponding feature points using Geometric Feature Correspondences.

14. The image processing apparatus according to claim 12 , wherein the reference image and the test image are images of a stereo pair.

15. The image processing apparatus according to claim 12 , wherein the at least one processor is further configured to:

determine a refined color mapping model from image colors of the reduced set of pairs of corresponding feature points for mapping image colors of the reduced set of pairs of corresponding feature points between the test image and the reference image;

apply the refined color mapping model to the test image to generate a refined color-corrected test image;

for each non-deleted pair of corresponding feature points, compare color values of a spatial area around a feature point of the non-deleted pair belonging to the reference image with color values of a corresponding spatial area around the other feature point of the non-deleted pair belonging to the refined color-corrected test image; and

when said comparison of color values related to a given non-deleted pair of corresponding feature points meets said threshold criteria, delete the given non-deleted pair of corresponding feature points to generate a further reduced set of pairs of corresponding feature points.

16. The image processing apparatus according to claim 12 , wherein the at least one processor is configured to determine Color Correspondences for determining the initial color mapping model.

17. The image processing apparatus according to claim 12 , wherein the at least one processor is configured to determine the threshold as a variance of an estimation error determined as a difference between a reference color and an estimated color.

18. The image processing apparatus according to claim 12 , wherein the at least one processor is configured to compare color values based on an absolute difference between mean color values of colors in said corresponding spatial areas for each color channel, and to delete a pair if said absolute difference exceeds a threshold for a majority of color channels.

19. The image processing apparatus according to claim 12 , wherein the initial color mapping model is a color transform.

20. The image processing apparatus according to claim 12 , wherein the initial color mapping model is a gamma-offset-gain-model.

21. The image processing apparatus according to claim 12 , wherein the at least one processor is also configured to perform said comparison of color values directly pixel to pixel in said corresponding spatial areas to decide whether pixels of each pixel to pixel comparison match together, and to delete a pair if a total number of pixel to pixel matches in the corresponding spatial areas of the given pair of corresponding feature points is below a threshold.

22. The method according to claim 12 , wherein the at least one processor is also configured to cluster colors of corresponding spatial areas of each pair, to compare color values based on a difference computed between colors of color clusters that correspond in said corresponding spatial areas and to delete a pair if the difference computed for the given pair of corresponding feature points exceeds a threshold.