Joint Motion and Disparity Fields Estimation for Stereoscopic Video Sequences

摘要

In this paper, we present a separate edge-preserving regularization scheme to calculate disparity fields for a stereoscopic image pair and a joint disparity and motion estimation algorithm for stereoscopic video sequences. We aim at using the block-based joint estimation algorithm to calculate the displacement fields for stereoscopic and multiview video coding. In the proposed separate regularization scheme, an edge-preserving cost function is proposed for matching, the Sobel edge values are incorporated into the cost function as edge-preserving weights. The optimal Lagrange multiplier is determined using the convex hull bisection algorithm under the rate-distortion theory. A fast algorithm is then proposed where the textured regions and the homogeneous regions of the images are identified and regularized differently. In the joint regularization scheme, we calculate the two motion fields and the two disparity fields for two successive image pairs simultaneously. The four fields are regularized iteratively under the stereo-motion consistency constraint. Results are compared with existing algorithms and the superior performance of the proposed methods is confirmed.