传统的基于全局优化的多视网立体匹配方法由于算法复杂度和内存容量的限制,难以处理高分辨率的图像,为此提出一种保细节的层次化多视图它体匹配方法.首先对每幅分割的图像进行结构微细度评估,为分割区域设定合适的匹配分辨率层;然后在图像的金字塔结构上:进行由粗到细的视差恢复:在一定的分辨率层上,对属于该分辨率层的图像区域进行全局优化的视差恢复,以保证精细结构区域在合适的分辨率上进行恢复;对已经处理过的图像区域进一步求精.此外,还提出了一个分块化的求解策略,可将大分辨率图像分块化匹配,从而能以较小的内存容量运行全局优化算法,突破了内存容量的制约.实验结果充分表明了文中方法的有效性,其在高效求解的同时又能保证精细结构的视差恢复.%Traditional global optimization based stereo matching methods generally have difficulties in processing high-resolution images, due to their high computational complexity and large memory space requirement. This paper proposes a novel detail-preserving hierarchical multi-view stereo matching method. In the proposed method, for each segmented image, we first evaluate the degree of fine structures and assign each segment with an appropriate resolution level for matching. Then, we perform a coarse-to-fine disparity estimation on the pyramid structure of the image. For each resolution level, we estimate the disparity with global optimization for the regions with corresponding level label,so that fine structures can be processed with appropriate resolutions. For regions processed in previous resolution levels, we just conduct a further disparity refinement. In addition, a block-based solving strategy is proposed, which divides the high-resolution image into multiple blocks, so that the global optimization algorithm can be performed with small memory space. The experimental results demonstrate that the proposed method is very efficient while preserving fine structures.
展开▼
