Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for detection of pixel correspondences in stereo imaging through the use of polarization data. One of the methods include: obtaining a first image from a first viewpoint of a stereo pair and a second image from a second viewpoint of the stereo pair; obtaining an angle of linear polarization (AOLP) map and a degree of linear polarization (DOLP) map having a field of view overlapping the first image and the second image; computing a surface normal map based on the AOLP map and the DOLP map; and detecting corresponding pixels between the first image and the second image by computing a disparity map that minimizes an energy function comprising a pixel matching cost term and a polarized smoothness regularization term computed based on pixels in a local neighborhood and based on the surface normal map.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for detection of pixel correspondences in stereo imaging through the use of polarization data. One of the methods include: obtaining a first image from a first viewpoint of a stereo pair and a second image from a second viewpoint of the stereo pair; obtaining an angle of linear polarization (AOLP) map and a degree of linear polarization (DOLP) map having a field of view overlapping the first image and the second image; computing a surface normal map based on the AOLP map and the DOLP map; and detecting corresponding pixels between the first image and the second image by computing a disparity map that minimizes an energy function comprising a pixel matching cost term and a polarized smoothness regularization term computed based on pixels in a local neighborhood and based on the surface normal map.