Scene adaptive filter design for improved stereo matching

1. A method comprising:

generating coefficients of a scene adaptive filter (SAF) based on differences between values of neighboring pixels in a representative two dimensional (2D) image; and

applying the SAF to a plurality of corresponding 2D images, wherein generating coefficients comprises applying each test filter of a plurality of test filters to the 2D image, computing a statistical measure of efficacy for each test filter based on absolute differences computed for each pixel of interest of a plurality of pixels of interest in the 2D image as filtered by the test filter, wherein the absolute differences for a pixel of interest are absolute differences between the pixel of interest and each pixel in a horizontal shift range centered on the pixel of interest; and selecting coefficients of a test filter of the plurality of test filters having a best statistical measure as the SAF coefficients.

2. The method of claim 1 , wherein generating coefficients further comprises generating the plurality of test filters based on a parameterized filter model, wherein each test filter is generated according to differing parameter values.

3. The method of claim 1 , wherein the plurality of pixels of interest includes every pixel in the 2D image.

4. The method of claim 1 , wherein generating coefficients comprises:

computing, for each pixel of interest of a plurality of pixels of interest in the 2D image, absolute differences between corresponding pixels in a base M×N window centered on the pixel of interest and comparison M×N windows, wherein each comparison M×N window is centered on a pixel in a horizontal shift range centered on the pixel of interest, and wherein M and N are dimensions of the SAF;

computing raw coefficient values for each coefficient position of the SAF filter by summing the absolute differences computed for a corresponding pixel location in each base M×N window; and

normalizing the raw coefficient values to generate final coefficient values.

5. The method of claim 4 , wherein the plurality of pixels of interest includes every pixel in the 2D image.

6. An apparatus comprising:

means for generating coefficients of a scene adaptive filter (SAF) based on differences between values of neighboring pixels in a representative two dimensional (2D) image; and

means for applying the SAF to a plurality of corresponding 2D images, wherein the means for generating coefficients comprises means for applying each test filter of a plurality of test filters to the 2D image, means for computing a statistical measure of efficacy for each test filter based on absolute differences computed for each pixel of interest of a plurality of pixels of interest in the 2D image as filtered by the test filter, wherein the absolute differences for a pixel of interest are absolute differences between the pixel of interest and each pixel in a horizontal shift range centered on the pixel of interest, and means for selecting coefficients of a test filter of the plurality of test filters having a best statistical measure as the SAF coefficients.

7. The apparatus of claim 6 , wherein the means for generating coefficients further comprises means for generating the plurality of test filters based on a parameterized filter model, wherein each test filter is generated according to differing parameter values.

8. The apparatus of claim 6 , wherein the plurality of pixels of interest includes every pixel in the 2D image.

9. The apparatus of claim 6 , wherein the means for generating coefficients comprises:

means for computing, for each pixel of interest of a plurality of pixels of interest in the 2D image, absolute differences between corresponding pixels in a base M×N window centered on the pixel of interest and comparison M×N windows, wherein each comparison M×N window is centered on a pixel in a horizontal shift range centered on the pixel of interest, and wherein M and N are dimensions of the SAF;

means for computing raw coefficient values for each coefficient position of the SAF filter by summing the absolute differences computed for a corresponding pixel location in each base M×N window; and

means for normalizing the raw coefficient values to generate final coefficient values.

10. The apparatus of claim 9 , wherein the plurality of pixels of interest includes every pixel in the 2D image.

11. A non-transitory computer readable medium storing software instructions that, when executed by a processor, cause the performance of a method comprising:

generating coefficients of a scene adaptive filter (SAF) based on differences between values of neighboring pixels in a representative two dimensional (2D) image; and

applying the SAF to a plurality of corresponding 2D images, wherein generating coefficients comprises applying each test filter of a plurality of test filters to the 2D image, computing a statistical measure of efficacy for each test filter based on absolute differences computed for each pixel of interest of a plurality of pixels of interest in the 2D image as filtered by the test filter, wherein the absolute differences for a pixel of interest are absolute differences between the pixel of interest and each pixel in a horizontal shift range centered on the pixel of interest, and selecting coefficients of a test filter of the plurality of test filters having a best statistical measure as the SAF coefficients.

12. The computer readable medium of claim 11 , wherein generating coefficients further comprises generating the plurality of test filters based on a parameterized filter model, wherein each test filter is generated according to differing parameter values.

13. The computer readable medium of claim 11 , wherein the plurality of pixels of interest includes every pixel in the 2D image.

14. The computer readable medium of claim 11 , wherein generating coefficients comprises:

computing, for each pixel of interest of a plurality of pixels of interest in the 2D image, absolute differences between corresponding pixels in a base M×N window centered on the pixel of interest and comparison M×N windows, wherein each comparison M×N window is centered on a pixel in a horizontal shift range centered on the pixel of interest, and wherein M and N are dimensions of the SAF;

computing raw coefficient values for each coefficient position of the SAF filter by summing the absolute differences computed for a corresponding pixel location in each base M×N window; and

normalizing the raw coefficient values to generate final coefficient values.

15. The computer readable medium of claim 14 , wherein the plurality of pixels of interest includes every pixel in the 2D image.