应用分层MRF/GRF模型的立体图像视差估计及分割

Hierarchical MRF/GRF Model Based Disparity Estimation and Segmentation for Stereo Images

视差估计与分割是立体图像编码及立体视觉匹配的核心问题 ,本文提出一种基于分层MRF/GRF模型和交叠块匹配 (HMOM)视差估计算法以及结合主动轮廓模型的视差分割提取算法 .该混合视差估计方法 ,可得到光滑准确 ,且具有清晰边缘的视差场 ;并便于用主动轮廓模型提取感兴趣对象 (OOI)的视差轮廓 .与通常的变尺寸块匹配(VSBM)相比 ,本算法得到的视差补偿图像的峰值信噪比可提高 2 .5dB左右 .

Disparity estimation and segmentation are vital tasks in stereo image coding and stereo vision matching. This paper proposes an algorithm for disparity estimation based on hierarchical MRF/CRF model and overlapped block matching (HMOM), and for disparity segmentation and extraction using active contour model. By combining the advantages of variable size block matching, overlapped block matching and MRF model, the HMOM scheme estimates a more smoother and accurate disparity field with lower computational complexity. In terms of PSNR of the disparity compensated image, the proposed disparity estimation algorithm achieves about 2.5 dB higher PSNR, as compared to conventional variable block matching. A smooth and consistent disparity field with sharp boundary can be obtained by pixel-wise disparity refinement on the HMOM disparity field, and the disparity contour of OOI (objects of interest) can be extracted by the active contour model. The resulting disparity field and corresponding disparity contours can further be used in stereo image coding and video object segmentation.