主权项 |
1. A method for objectively evaluating quality of a stereo image comprising following steps of:
{circle around (1)} defining an original stereo image which is utilized for referring in an objective quality evaluation of the stereo image as a reference stereo image, a left-view of the reference stereo image is denoted as Il, a right-view of the reference stereo image is denoted as Ir a cyclopean image of the reference stereo image formed in a human visual system is denoted as C; defining a distortion image of the reference stereo image which is used for testing in the objective quality evaluation of the stereo image as a testing stereo image, a left-view of the testing stereo image is denoted as Îl, a right-view of the testing stereo image is denoted as Îr, a cyclopean image of the testing stereo image formed in the human visual system is denoted as Ĉ, wherein sizes of the left-view of the reference stereo image Il, the right-view of the reference stereo image Ir, the left-view of the testing stereo image Îl, and the right-view of the testing stereo image Îr, are all denoted as W×H, wherein W refers to a width thereof, and H refers to a height thereof; {circle around (2)} classifying the cyclopean image Ĉ of the testing stereo image formed in the human visual system into three areas, wherein: a first area thereof is an occlusion area, which is denoted as {circumflex over (R)}occ, information of the occlusion area {circumflex over (R)}occ exists only in the left-view or the right-view of the testing stereo image; a second area thereof is a binocular suppression area, which is denoted as {circumflex over (R)}bs, information of the binocular suppression area {circumflex over (R)}bs is binocular information having great image content difference or great disparity between the left-view and the right-view of the testing stereo image; and a third area thereof is a binocular fusion area, which is denoted as {circumflex over (R)}bf, information of the binocular fusion area {circumflex over (R)}bf is binocular information having similar image content and small disparity between the left-view and the right-view of the testing stereo image; {circle around (3)} dividing the cyclopean image Ĉ of the testing stereo image formed in the human visual system into a plurality of image blocks which are non-overlapped with each other and have a size of k×k, classifying all of the image blocks into three types: an occlusion block, a binocular suppression block and a binocular fusion block, updating {circumflex over (R)}occ, {circumflex over (R)}bs and {circumflex over (R)}bf according to types of the image blocks, so as to obtain an updated occlusion area {circumflex over (R)}occ′, an updated binocular suppression area {circumflex over (R)}bs′ and an updated binocular fusion area {circumflex over (R)}bf′, in the cyclopean image C of the reference stereo image formed in the human visual system, obtaining an area Rocc′ corresponding to a position of {circumflex over (R)}occ′, an area Rbs′ corresponding to a position of {circumflex over (R)}bs′, and an area Rbf′ corresponding to a position of {circumflex over (R)}bf′, processing singular value decomposition respectively on all of the image blocks having a size of k×k in the area {circumflex over (R)}occ′, an area Rbsl′ corresponding to left-view of {circumflex over (R)}bs′, an area {circumflex over (R)}bsr′ corresponding to right-view of {circumflex over (R)}bs′, an area {circumflex over (R)}bfl′ corresponding to left-view of {circumflex over (R)}bf′, an area {circumflex over (R)}bfr′ corresponding to right-view of {circumflex over (R)}bf′, the area Rocc′, an area Rbsl′ corresponding to left-view of Rbs′, an area Rbsr′ corresponding to right-view of Rbs′, an area Rbfl′ corresponding to left-view of Rbf′, and an area Rbfr′ corresponding to right-view of Rbf′, in such a manner that singular values respectively corresponding to each image block are obtained, wherein k>1; {circle around (4)} obtaining an overall distortion degree Qocc of {circumflex over (R)}occ′ according to singular value distances between all of the image blocks which are in corresponding positions in {circumflex over (R)}occ′ and {circumflex over (R)}occ′ and have a size of k×k, obtaining an overall distortion degree Qbs of {circumflex over (R)}bs′ according to singular value distances between all of the image blocks which are in corresponding positions in {circumflex over (R)}bsl′ and Rb,sl′ and have a size of k×k, and singular value distances between all of the image blocks which are in corresponding positions in {circumflex over (R)}bsr′ and Rbsr′ and have a size of k×k, and obtaining an overall distortion degree Qbf of {circumflex over (R)}bf′ according to singular value distances between all of the image blocks which are in corresponding positions in {circumflex over (R)}bfl′ and Rbfl′ and have a size of k×k, and singular value distances between all of the image blocks which are in corresponding positions of {circumflex over (R)}bfr′ and Rbfr′ and have a size of k×k; and {circle around (5)} according to the overall distortion degree Qocc of {circumflex over (R)}occ′, the overall distortion degree Qbs of {circumflex over (R)}bs′ and the overall distortion degree Qbf of {circumflex over (R)}bf′, calculating an overall distortion degree Q of the testing stereo image by an equation of linear regression, wherein Q=a×Qocc+b×Qbs+c×Qbf, wherein a, b and c are all coefficients and satisfy: {a+b+c=10≤a≤10≤b≤10≤c≤1. |