基于稳态匹配概率和边缘感知视差传播的立体匹配算法研究

Stereo matching algorithm based on steady-state matching probability and edge-aware disparity propagation

提出一种由稳态匹配概率(steady-state matching probability,SSMP)和边缘感知视差传播(edge-aware disparity propagation,EDP)相结合的立体匹配算法。首先,基于SSMP求取初始视差图,通过一致性检测滤除错误视差点。然后,利用EDP算法、测地距滤波等策略对一致视差点进行计算,得到可靠匹配成本。最后,由可靠匹配成本通过胜者为王(winner takes all,WTA)算法逐像素优化填充到对应的不一致视差点,从而使视差图更精确。数值实验表明,该方法利用了正确的视差值来估计因遮蔽、噪声等原因产生的错误像素点的视差值,能有效提高匹配精度并得到较丰富的视差图细节,且结构简单,计算复杂度低。

A stereo matching algorithm based on steady-state matching probability（ SSMP） and edge-aware disparity propagation（ EDP） is proposed. Firstly,based on the initial parallax image with SSMP,the inconsistent pixels with the left-right check are filtered out. Secondly,EDP and the geodesic filter with the consistent disparities are used to obtain the reliable cost volume. Finally,the inconsistent pixels are filled into the disparity map with disparity values selected by winner take all（ WTA）. The numerical experiment shows that the method can effectively improve the matching accuracy by using the right disparity values to evaluate wrong disparity values of other pixels generated by the occlusive region and noises. So it can obtain disparity map with more details and the filtering has lower complexity.