Method and apparatus for analysing an array of pixel-to-pixel dissimilarity values by combining outputs of partial filters in a non-linear operation
View Patent ↗In for example the assignment of motion vectors, an array of pixel-to-pixel dissimilarity values is analyzed to identify a pixel which has a low pixel-to-pixel dissimilarity value and which has neighboring pixels which have a low pixel-to-pixel dissimilarity value. The pixel-to-pixel dissimilarity values are filtered with a filter aperture decomposed into two or more sectors with partial filters applied respectively to each sector. Outputs of the partial filters are combining by a non-linear operation, for example taking the minimum from diametrically opposed sectors.
1. In a video processor, a method of analysing an array of pixel-to-pixel dissimilarity values to identify a pixel which has a low pixel-to-pixel dissimilarity value and which has neighbouring pixels which have a low pixel-to-pixel dissimilarity value; the method comprising the steps of:
receiving an array of pixel-to-pixel dissimilarity values, each pixel-to-pixel dissimilarity value representing a difference between the value of a pixel in a first spatial pixel array representing a first image and the value of a corresponding pixel in a second spatial pixel array representing a second image;
providing a filter having a filter aperture which defines those pixel-to-pixel dissimilarity values on which the filter operates and which is decomposed into two or more filter windows which define respective subsets of those pixel-to-pixel dissimilarity values on which the filter operates, wherein said filter comprises a plurality of partial filters applied respectively to the windows of the filter aperture;
filtering the pixel-to-pixel dissimilarity values in each filter window with the respective partial filter to provide an output for each partial filter; and
combining the respective outputs of said partial filters by a non-linear operation, in which the non-linear combination process includes taking minimum values of partial-filter outputs from pairs of filter windows that are diametrically opposite each other in the filter aperture.
2. A method according to claim 1 in which each dissimilarity value is a rectified field difference between fields displaced through motion compensation.
3. A method according to claim 1 in which the partial filters operate on radial line segments.
4. A method according to claim 1 in which the partial filters operate on sectors of a circle in the filter aperture.
5. A method according to claim 1 in which the partial filtering operation is a rank-order operation.
6. A method according to claim 1 in which the partial filtering operation is an averaging operation.
7. A method according to claim 1 in which the minimum values from said pairs of filter windows are processed by a rank-order operation.
8. A method according to claim 1 in which the minimum values from said pairs of filter windows are processed by an averaging operation.
9. Video sequence processing apparatus for analysing motion-compensated pixel-to-pixel dissimilarity values, the apparatus comprising
a spatial filter configured to operate on motion-compensated pixel-to-pixel dissimilarity values in which the filter aperture which defines those values on which the filter operates is decomposed into two or more filter windows which define respective subsets of those pixel-to-pixel dissimilarity values on which the filter operates and the filter comprises a plurality of partial filters applied respectively to each filter window; and
a combiner for combining the respective outputs of said partial filters by a non-linear operation in which the combiner operates to take minimum values of partial-filter outputs from pairs of filter windows that are diametrically opposite each other in the filter aperture.
10. Apparatus according to claim 9 in which the partial filters operate on sectors a circle in of the filter aperture.
11. Apparatus according to claim 9 in which the partial filtering operation is a rank-order operation.
12. Apparatus according to claim 9 in which the partial filtering operation is an averaging operation.
13. Apparatus according to claim 9 in which the minimum values from said pairs of filter windows are processed by a rank-order operation.
14. Apparatus according to claim 9 in which the minimum values from said pairs of filter windows are processed by an averaging operation.
15. A method of video processing in a video processor, comprising the steps of:
deriving a candidate predictor representing a displacement between first and second image regions of the video sequence, each image being formed of pixels;
using the predictor to project from the first image region to the second image region a current pixel and a plurality of pixels neighbouring the current pixel;
providing for each pixel a dissimilarity value indicative of the dissimilarity between the pixel in the second image and the pixel projected from the first image; and
analysing the plurality of dissimilarity values to derive an error value for the candidate predictor;
wherein the step of analysing the plurality of dissimilarity values comprises:
dividing the pixels neighbouring the current pixel into at least two diametrically disposed sectors:
deriving a high error value for the predictor when the pixels in all sectors have high dissimilarity values; and
deriving a low error value for the predictor when the current pixel and the pixels of at least one sector have low dissimilarity values.
16. A non-transitory computer program product adapted to cause programmable apparatus to implement a method of analysing an array of pixel-to-pixel dissimilarity values to identify a pixel which has a low pixel-to-pixel dissimilarity value and which has neighbouring pixels which have a low pixel-to-pixel dissimilarity value; the method comprising the steps of filtering the pixel-to-pixel dissimilarity values with a filter having a filter aperture which defines those pixel-to-pixel dissimilarity values on which the filter operates decomposed into two or more filter windows which define respective subsets of those pixel-to-pixel dissimilarity values on which the filter operates with a plurality of partial filters applied respectively to each filter window; and combining the outputs of said partial filters by a non-linear operation in which the non-linear combination process includes taking minimum values of partial-filter outputs from pairs of filter windows that are diametrically opposite each other in the filter aperture.
17. A non-transitory computer program product according to claim 16 , wherein the two or more regions comprise sectors a circle.