基于最优斜面参数估计的局部立体匹配算法

针对传统局部匹配算法在斜面场景匹配中所表现出的阶梯效应，提出了一种基于最优斜面参数估计的局部立体匹配算法。该算法首先为每一个像素随机地分配一组斜面参数，然后以新的斜面参数所定义的支撑域下当前像素的匹配代价是否减小为准则，迭代地进行斜面参数的邻域传播-单点优化过程，并最终使得计算结果收敛到最优斜面，同时估计得到稠密的亚像素级视差。通过对典型斜面场景图像和Middlebury 标准测试图像对的匹配实验表明，文中算法在将对普通场景的匹配效果保持在当前先进水平的同时，对斜面场景的匹配消除了阶梯效应，且匹配率代表了局部匹配的先进水平。

Best slant-plane estimation based stereo matching algorithm

A novel stereo matching algorithm based on best slant-plane estimation was proposed in this paper in the purpose of eliminating stair-casing which showed up frequently in the slant scene matching process where the window-based matching algorithm was used. In this procedure, a slant parameter vector was randomly attributed to every pixel in the reference image firstly, then, those vectors were iteratively propagated between neighbor pixels followed by a recursively slant-plane parameter refinement process for each pixels in the principle of whether a lower cost could be got under the new slant-plane parameter vectors, until the parameter vectors were converged to the best slant-plane parameter vectors while a sub-pixel disparity was got for each pixel in the reference image. Experiment results indicate the effectiveness of the algorithm, the performance of the algorithm on the slant scene is ranked on top of those state-of-art algorithm which is relatively close to the algorithm proposed here, while the performance on the normal scene is comparable with the state-of-art algorithm.