基于梯度和MRF模型的视差估计算法

Disparity estimation based on the gradient and MRF model

提出一种基于梯度和MRF（Markov Random Field）模型的视差估计及优化算法．采用图像灰度和梯度加权联合的方法进行块匹配运算，获得初始视差场．然后根据顺序匹配准则对该视差场进行交叉块检测，并运用基于MRF模型的因果预测对误匹配块进行迭代校正，最终获得较为精确平滑的视差场．实验表明，与传统块匹配法相比，该算法生成的视差场能够将预测图像峰值信噪比提高1．2～1．8dB，且运算时问在1s以内．

This paper presents a novel disparity estimation algorithm based on the gradient and Markov Random Field （MRF） model. First, the block matching algorithm combining gray and gradient information is adopted to obtain an initial disparity field. Second, an order matching constraint is applied to detect cross regions in the disparity-map. Finally, the erroneously matched blocks are corrected iteratively by MRF-based causality prediction to achieve a more accurate disparity field. Experimental results show that the proposed algorithm achieves a PSNR gain（about 1.2-1.8 dB） as compared to the conventional block-based method and its calculating time is less than 1 s.