Method of filtering a video sequence image from spurious motion effects

• US 8,150,181 B2
• Filed: 11/17/2008
• Issued: 04/03/2012
• Est. Priority Date: 11/17/2008
• Status: Active Grant

• Current Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Original Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Inventors: Bruna, Arcangelo Ranieri, Battiato, Sebastiano, Puglisi, Giovanni
• Primary Examiner(s): Do, Anh Hong
• Application Number:
  US12/272,591
• Publication Number:
  US 20100124379A1
• Time in Patent Office: 1,233 Days
• Field of Search: 382/236, 382/254, 382/239, 382/154, 345/603, 345/530, 345/555, 375/240.02, 375/240.12, 375/E7.093, 375/E7.094, 348/416.1, 709/247
• US Class Current: 382/236
• CPC Class Codes: H04N 19/51 Motion estimation or motion compensation
• CPC Class Codes: H04N 19/80 Details of filtering operations specially adapted for video compression, e.g. for pixel interpolation H04N19/635, H04N19/86 take precedence
• CPC Class Codes: H04N 19/86 involving reduction of coding artifacts, e.g. of blockiness

Abstract

According to the novel method, roto-translational and zooming parameters describing spurious motion effects are determined by exploiting any of the many block matching algorithms commonly used for motion estimation for calculating a motion vector for all or for a selected number of blocks of pixels of the current frame that is processed. Some of the so calculated motion vectors are not taken into account for estimating spurious motion effects. The roto-translational and zooming parameters describing what is considered to be spurious global motion between a current frame and the precedent frame of the sequence, are calculated by processing the selected motion vectors of blocks of pixels of the frame through a recursive procedure that includes computing error values and readjusting the roto-translational and zooming parameters based on the error values.

Claims

1. A method, comprising:

filtering, using one or more processors, a current image of an input video sequence from spurious motion effects, the filtering including;

calculating motion vectors for blocks of pixels of the current image of the video sequence with a block matching algorithm carried out on corresponding blocks of pixels of the current image and of a preceding image that precedes the current image in the sequence;

selecting a subset of said motion vectors by deselecting;

motion vectors calculated from blocks of the current image dissimilar from corresponding blocks of the preceding image,

motion vectors strongly different from motion vectors of surrounding blocks, and

motion vectors associated with pixels of homogeneous areas of the current image;

calculating roto-translational and zooming parameters describing spurious global motion between the current image and the preceding image, by processing the motion vectors of said subset through the following recursive procedure;

calculating, for each current motion vector of said subset, a corresponding expected motion vector estimated in function of roto-translational and zooming parameters relative to the preceding image,

calculating for each current motion vector of said subset a respective error value in function of the motion vector of the corresponding block of the preceding image and the corresponding expected motion vector,

comparing said error values with a first threshold and storing in a memory the current motion vectors of said subset having error values that are smaller than said first threshold and deleting from said memory motion vectors previously stored therein having error values that are larger than said first threshold, and

calculating said roto-translational and zooming parameters for the current image in function of motion vectors stored in said memory and of current motion vectors of said subset; and

generating a filtered output image from said current image of the input video sequence by compensating spurious motion effects described by said roto-translational and zooming parameters.

2. The method of claim 1, wherein said first threshold is determined for each image such to store in said memory a pre-established number of motion vectors.

3. The method of claim 1, wherein deselecting motion vectors calculated from blocks of the current image dissimilar from the corresponding blocks of the preceding image includes:

calculating, for a block of pixels of the current image, a sum of absolute values of differences between pixels of said block and of a corresponding block of the preceding image, and

identifying as dissimilar the two blocks if said sum exceeds a second threshold.

4. The method of claim 1, wherein deselecting motion vectors strongly different from motion vectors of surrounding blocks includes:

calculating a sum of absolute values of differences between a component of the considered motion vector and corresponding components of surrounding motion vectors, and

identifying the considered motion vector as strongly different from motion vectors of surrounding blocks if said sum exceeds a second threshold.

5. The method of claim 1, wherein deselecting motion vectors associated with pixels of homogeneous areas of the current image includes identifying a considered block of pixels of the current image as depicting an inhomogeneous area through the following operations:

calculating, for the considered block of pixels of the current image, a horizontal inhomogeneity parameter as a sum of absolute values of differences between corresponding pixels of said considered block and of an adjacent block in horizontal direction,

calculating, for the considered block of pixels of the current image, a vertical inhomogeneity parameter as a sum of absolute values of differences between corresponding pixels of said considered block and of an adjacent block in vertical direction,

comparing said horizontal and vertical inhomogeneity parameters with respective second and third thresholds, and

identifying said considered block of pixels as depicting a homogeneous area if either or both said second and third thresholds are not exceeded.

6. The method of claim 5, wherein:

deselecting motion vectors calculated from blocks of the current image dissimilar from the corresponding blocks of the preceding image includes;

calculating, for a block of pixels of the current image, a sum of absolute values of differences between pixels of said block and of a corresponding block of the preceding image, and

identifying as dissimilar the two blocks if said sum exceeds a fourth threshold;

deselecting motion vectors strongly different from motion vectors of surrounding blocks includes identifying a considered motion vector as being strongly different from motion vectors of surrounding blocks through the following operations;

calculating a sum of a absolute values of a differences between a components of the considered motion vector and corresponding components of surrounding motion vectors, and

identifying the considered motion vector as strongly different from motion vectors of surrounding blocks if said sum exceeds a fifth threshold; and

said thresholds from second to fifth are determined such that the number of motion vectors of said subset is a pre-established percentage of the calculated motion vectors.

7. The method of claim 1, further comprising selecting for the current image a set of significant blocks, wherein calculating motion vectors is carried out by processing only said significant blocks.

8. The method of claim 7, wherein said set of significant blocks comprises randomly chosen blocks and blocks of the current image corresponding to significant blocks for the preceding image.

9. The method of claim 8, wherein said set of significant blocks for the current image is determined through the following steps:

processing said blocks of the current image corresponding to significant blocks for the preceding image with a morphological operator, identifying a first intermediate set of blocks of the current image;

identifying a second intermediate set of blocks of the current image by deselecting a number of blocks of said first intermediate set in a spatially uniform fashion; and

identifying said set of significant blocks for the current image as the set composed of the blocks of said second intermediate set and of a number of randomly chosen groups of four adjacent blocks of the current image.

10. The method of claim 1, wherein said error value being either a length of a vector difference between a current motion vector and a corresponding expected motion vector, or a cosine of an angle comprised between the current motion vector and the corresponding expected motion vector.

11. The method of claim 10, comprising:

calculating for each current motion vector of said subset two respective error values, a first error value being a length of a vector difference between a motion vector and a corresponding expected motion vector and a second error value being a cosine of an angle comprised between the motion vector and the corresponding expected motion vector;

comparing said first and second error values with said first threshold and with a second threshold, respectively, and storing in the memory motion vectors of said subset having respective error values that are at the same time smaller than said first threshold and said second threshold and deleting from said memory motion vectors having respective error values the are larger than said first threshold or said second threshold.

12. The method of claim 11, wherein said roto-translational and zooming parameters are calculated through the following steps:

calculating provisional values of said roto-translational and zooming parameters by applying a least squares algorithm only on the motion vectors stored in said memory;

for each of said current motion vectors of said subset having respective error values that are smaller than said first and second thresholds, recalculating said respective error values using said provisional values of said roto-translational and zooming parameters and comparing them again with said first and second thresholds; and

selecting among said subset of current motion vectors, based upon the comparing of the recalculated error values with the first and second thresholds.

13. The method of claim 1, comprising selecting current blocks of pixels used for calculating motion vectors as blocks of the current image corresponding to the blocks of pixels of the preceding image the motion vectors of which have been used for calculating said roto-translational and zooming parameters.

14. The method of claim 13, further comprising determining whether current blocks of pixels eventually to be processed are substantially regular and, in the affirmative case, processing said substantially regular current blocks even if they do not correspond to any of said blocks of pixels of the preceding image.

15. The method of claim 14, wherein said determining step is carried out through a fuzzy logic algorithm.

16. The method of claim 15, wherein said fuzzy logic algorithm comprises:

calculating, for each current block of pixels;

a first parameter distance according to the following formula

17. The method of claim 16, wherein each of said antecedents is compared with a respective pair of discrimination levels, said current block of pixels being discriminated upon said comparisons.

18. The method of claim 17, comprising determining said discrimination levels by:

procuring a first set of test blocks of pixels showing a periodic pattern and a second set of test blocks of pixels showing an aperiodic pattern;

calculating a mean Euclidean distance and an average number for each of said test blocks;

fixing said discrimination levels through a genetic algorithm processing the mean Euclidean distance and average number, such to minimize a probability of erroneous discrimination.

19. A method, comprising:

generating an enhanced output image by enhancing a digital input image using one or more processors, the enhancing including;

subdividing the input image into blocks of pixels;

discriminating substantially regular from substantially irregular blocks of pixels by calculating, for each current block of pixels;

a first parameter distance according to the following formula

20. The method of claim 19, wherein the enhancing includes comparing of said antecedents with discrimination levels, respectively, and said discriminating includes discriminating said current block of pixels based on said comparing.

21. The method of claim 20, comprising determining said discrimination levels by:

procuring a first set of test blocks of pixels showing a periodic pattern and a second set of test blocks of pixels showing an aperiodic pattern;

calculating a mean Euclidean distance and an average number for each of said test blocks;

fixing said discrimination levels through a genetic algorithm processing the mean Euclidean distance and average number, such to minimize a probability of erroneous discrimination.

22. A non-transitory computer-readable medium comprising contents that cause a computing device to implement a method that includes:

filtering a current image of an input video sequence from spurious motion effects, the filtering including;

calculating motion vectors for blocks of pixels of the current image of the video sequence with a block matching algorithm carried out on corresponding blocks of pixels of the current image and of a preceding image that precedes the current image in the sequence;

selecting a subset of said motion vectors by deselecting;

motion vectors calculated from blocks of the current image dissimilar from corresponding blocks of the preceding image,

motion vectors strongly different from motion vectors of surrounding blocks, and

motion vectors associated with pixels of homogeneous areas of the current image;

calculating roto-translational and zooming parameters describing spurious global motion between the current image and the preceding image, by processing the motion vectors of said subset through the following recursive procedure;

calculating, for each current motion vector of said subset, a corresponding expected motion vector estimated in function of roto-translational and zooming parameters relative to the preceding image,

calculating for each current motion vector of said subset a respective error value in function of the motion vector of the corresponding block of the preceding image and the corresponding expected motion vector,

comparing said error values with a first threshold and storing in a memory the current motion vectors of said subset having error values that are smaller than said first threshold and deleting from said memory motion vectors previously stored therein having error values that are larger than said first threshold, and

calculating said roto-translational and zooming parameters for the current image in function of motion vectors stored in said memory and of current motion vectors of said subset; and

generating a filtered output image from said current image of the input video sequence by compensating spurious motion effects described by said roto-translational and zooming parameters.

23. The non-transitory computer-readable medium of claim 22, wherein the method includes selecting current blocks of pixels used for calculating motion vectors as blocks of the current image corresponding to the blocks of pixels of the preceding image the motion vectors of which have been used for calculating said roto-translational and zooming parameters.

24. The non-transitory computer-readable medium of claim 23, wherein the method includes determining whether current blocks of pixels eventually to be processed are substantially regular and, in the affirmative case, processing said substantially regular current blocks even if they do not correspond to any of said blocks of pixels of the preceding image.

25. The non-transitory computer-readable medium of claim 24, wherein said determining comprises:

calculating, for each current block of pixels;

a first parameter distance according to the following formula

26. A device, comprising:

filtering means for filtering a current image of an input video sequence from spurious motion effects, the filtering including;

calculating motion vectors for blocks of pixels of the current image of the video sequence with a block matching algorithm carried out on corresponding blocks of pixels of the current image and of a preceding image that precedes the current image in the sequence;

selecting a subset of said motion vectors by deselecting;

motion vectors calculated from blocks of the current image dissimilar from corresponding blocks of the preceding image,

motion vectors strongly different from motion vectors of surrounding blocks, and

motion vectors associated with pixels of homogeneous areas of the current image;

calculating roto-translational and zooming parameters describing spurious global motion between the current image and the preceding image, by processing the motion vectors of said subset through the following recursive procedure;

calculating, for each current motion vector of said subset, a corresponding expected motion vector estimated in function of roto-translational and zooming parameters relative to the preceding image,

calculating for each current motion vector of said subset a respective error value in function of the motion vector of the corresponding block of the preceding image and the corresponding expected motion vector,

comparing said error values with a first threshold and storing in a memory the current motion vectors of said subset having error values that are smaller than said first threshold and deleting from said memory motion vectors previously stored therein having error values that are larger than said first threshold, and

calculating said roto-translational and zooming parameters for the current image in function of motion vectors stored in said memory and of current motion vectors of said subset; and

an image generator structured to generate a filtered output image from said current image of the input video sequence by compensating spurious motion effects described by said roto-translational and zooming parameters.

27. The device of claim 26, wherein the filtering means includes a fuzzy classifier configured to determine whether current blocks of pixels eventually to be processed are substantially regular and, in the affirmative case, processing said substantially regular current blocks even if they do not correspond to any of said blocks of pixels of the preceding image.

28. The device of claim 27, wherein said fuzzy classifer comprises:

means for calculating, for each current block of pixels;

a first parameter distance according to the following formula