论文部分内容阅读
基于多视点视图深度特征,提出一种通过简单块匹配运算划分多视点视图区域并估计区域视差的算法。首先基于深度对象的概念确定图像中具有不同深度的区域数量以及这些区域对应的区域视差,再根据误差最小化准则初步确定每个图像块所属区域。当区域中图像块数量小于某个阈值时,采用区域合并算法将该区域中的每个图像块合并到与它的视差最为接近的其它图像区域,通过迭代形成最终的有效图像区域划分。实验表明,该算法能够以图像块为基本单元有效地划分各深度层区域,并准确估计对应的区域视差。
Based on the multi-view depth features, this paper proposes an algorithm to divide the multi-view area and estimate the area disparity by simple block matching. First, based on the concept of depth objects, the number of regions with different depths in the image and the corresponding regional disparity in these regions are determined. Then, the region to which each image block belongs is preliminarily determined according to the error minimization criterion. When the number of image blocks in an area is less than a certain threshold, each image block in the area is merged into other image areas with the closest parallax by using the area combination algorithm, and the final effective image area division is formed through iteration. Experimental results show that this algorithm can effectively divide each depth layer by using image block as the basic unit, and accurately estimate the corresponding area disparity.