Epipolar geometry estimation and its application to image coding

An Epipolar Geometry-Based Fast Disparity Estimation Algorithm for Multiview Image and Video Coding

“…Given the centroid coordinates of the current macroblock to be predicted in , the corresponding epipolar line equation in can be computed by multiplying by the homogeneous centroid coordinates using (1). We propose this centroid-based epipolar line calculation in that the resulting epipolar line computed from the centroid, unlike the top-left corner of the macroblock [17], can (4) In addition, since it is well known in the correspondence problem that a larger matching window is desirable to achieve the reliable matching, we thus propose applying (4) only to the disparity search at the macroblock level, i.e., for a block-size of 16 16. Although the fine-grained block disparity search for submacroblocks are supported in a few recent video coding standards, e.g., MPEG-4 [28] and H.264/AVC [5], we only transform MPSC to obtain OPESC for the macroblock level disparity search, so that the prediction outliers from small matching windows or submacroblocks can be kept away from destroying the smooth disparity field.…”

“…Though epipolar geometry provides a geometrical constraint for the correspondences, limiting the search range to the epipolar line is not an optimal way for interviewpoint coding, because video coding does not target at finding true correspondences, but rather reference blocks that minimize the coding cost. As a consequence, there is only considerably minor improvement in the multiview image coding efficiency based on this approach [17], and this does not warrant the extra increase in the decoding complexity and the changes of bit-stream syntax. In addition to this work, a view synthesis prediction technique [18], [19] has recently been proposed to improve the interviewpoint coding efficiency by exploiting the known camera geometrical parameters.…”

“…However, so far, little effort has been made to integrate this theoretical principle from stereo vision with the emerging practical multiview image/video encoders. Among these very limited previous works, Hata and Etoh [17] proposed to search the best block match only along the one-dimensional (1-D) epipolar line, and the resultant displacement along the epipolar line is encoded to replace the conventional two-dimensional (2-D) motion vectors. Though epipolar geometry provides a geometrical constraint for the correspondences, limiting the search range to the epipolar line is not an optimal way for interviewpoint coding, because video coding does not target at finding true correspondences, but rather reference blocks that minimize the coding cost.…”

Object-Based Coding for Plenoptic Videos

“…Another related prediction method is to employ epipolar geometry between adjacent camera views to facilitate inter-viewpoint prediction [8], [9]. In principle, if the true depth value of an image pixel is known and the surface is Lambertian, its positions in adjacent calibrated cameras are known.…”

“…Since the number of variables involved is not very large (around 20 or less), the complexity in solving this problem is relatively low. Let {R * m,n |1 ≤ m ≤ M, 1 ≤ n ≤ N} be the optimal solution of the problem in (6)- (8) …”