基于三角形网格的立体视差估计和遮挡点检测混合算法

A mesh-based algorithm for disparity estimation and occlusion points detection

本文提出一种新的用于立体图像编码的视差估计和遮挡点检测混合算法。其中的视差估计方法利用极线约束条件,在缩小搜索范围的同时提高了视差估计的准确性。遮挡点检测方法仅使用了匹配点唯一性约束和视差梯度限制这两个基本条件,降低了算法的复杂度。整个算法利用DT(Dalaunaytriangulation)网格这一数学工具把散乱的点结合起来进行处理,使算法在实现方面更加简单化。本文算法首先对立体图像对中的左图像进行DT网格剖分,把各三角形的顶点作为“特征点”在右图像中寻找它们的匹配点。然后利用匹配点唯一性条件提取出其中一些顶点进行遮挡检测。实验结果表明,本文算法对“特征点”的视差估计比较准确,也能较为准确地检测出其中的遮挡点。借助DT网格在图像编码方面的优势,本文算法可以方便地用于立体图像编码。

This paper proposes a novel cooperative algorithm for disparity estimation and occlusion points detection, which is adaptive to stereo image coding. The method for disparity estimation uses the principle of epipolar line constraint, which improves the veracity of estimation with reduced searching range. The method for occlusion points detection uses two basic principles which are the uniqueness assumption and the disparity gradient limit. Using these two simple principles reduces complexity of algorithm. DT meshes combine these disheveled points into a complete unite in this algorithm, which reduces the difficulty of implement. The algorithm first describes the left image with DT mesh. As feature points, the vertexes of triangulations are matched in the right image for finding their corresponding points. Then some points are selected by uniqueness assumption to detect occlusion. Experimental results show the disparity estimation is accurate. And it can detect occlusion points preferably. The algorithm can be applied to stereo image coding because DT mesh has some advantages on image coding.