基于AD-CESUS联和测度的立体匹配算法

Stereo-Matching Algorithm Based on AD-Census Joint Measure

双目图像深度估计是许多现代立体视觉技术的重要基础。由于受到光线、纹理结构变化,前后遮挡,图像噪声等因素的影响,基于单特征的匹配算法缺乏鲁棒性。本文将基于像素点的AD测度函数与基于区域的Census测度函数,依据匹配置信程度实现自适应加权融合,形成联和测度函数。该联和测度函数可以将AD的单调性与Census的区域性有效结合,提升立体匹配算法的鲁棒性。通过实验测试,证明采用该联测度函数可以有效提高局部和全局匹配算法的匹配准确度,尤其是局部匹配算法。

Depth estimation for binocular images is an important foundation of many current stereo vision application technologies. Depth estimation based on single image feature is lack of robustness, due to the affection of such factors as variances in texture and light, occlusion, image noise, etc. In this paper, the AD similarity measure based on pixels and the Census measure based on regions are weighted fusion according to their stereo - matching confidence, and formed a joint measure function. Monotonicity feature of the AD measure and region feature of the Census measure are effectively combined together,which can help to ascend the robustness of stereo - matching algorithms. Experimental results demonstrate the effectiveness of the joint measure function by used in global method and local method, especially in the local stereo - matching algorithm.