为了确定同一场景在不同视点下图像点的对应关系,提出一种基于光照不变特征和改进型支持权重核的快速立体匹配。首先,稠密构造光照不变局部特征计算初始匹配代价;基于三次立方Tri—Cube权重核双通聚合得到稳健的匹配代价;采用优胜者全选法选择初始视差。然后,利用基于多方向邻域Tri—Cube权重核的双边滤波、遮挡检测和误码掩盖等策略实现视差智能修复。实验表明,该方法能产生较为满意的稠密视差图,计算费用低,能提高对光照变化的适应能力。
A fast stereo matching method based on illumination invariant features and modified support weights is proposed to determine the correspondences in different images of the same scene taken from different viewpoints. Firstly, illumination invariant features are constructed densely for the stereo pairs and initial matching costs are calculated from the invariant features; robust matching costs are obtained based on two-pass aggregation with tri-cube weight kernels; and initial disparities are selected using Winner-Takes-All optimization method. Secondly, intelligent disparity inpainting procedures are in turn implemented or executed with modified bilateral filtering based on multi-directional neighbor tri-cube weight kernels, occlusion detection and disparity error concealment. Experiments indicate that this technique with low computational burden, can produce comparably satisfactory dense disparity map, and can improve the adaptability to illumination variations.