In stereo matching, homogeneous areas, depth continuity areas, and occluded areas need more attention. Many methods try to handle pixels in homogeneous areas by propagating supports. As a result, pixels in homogeneous areas get assigned disparities inferred from the disparities of neighboring pixels. However, at the same time, pixels in depth discontinuity areas get supports from different depths and/or from occluded pixels, and resultant disparity maps are easy to be blurred. To resolve this problem, we propose a non-linear diffusion-based support aggregation method. Supports are iteratively aggregated with the support-weights, while adjusting the support-weights according to disparities to prevent incorrect supports from different depths and/or occluded pixels. As a result, the proposed method yields good results not only in homogeneous areas but also in depth discontinuity areas as the iteration goes on without the critical degradation of performance.
Support Aggregation via Non-linear Diffusion with Disparity-Dependent Support-Weights for Stereo Matching
조회 수 3568 댓글 0
|저 자||Kukjin Yoon, Yekeun Jeong, In So Kweon|
|학 회||Asian Conference on Computer Vision (ACCV), oral presentation|