应用于多视点视频编码的基于深度特征的图像区域分割与区域视差估计

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.

Depth Characteristic-Based Image Region Partition and Regional Disparity Estimation for Multi-View Video Coding

Based on depth characteristic of multi-view images, an image region partition and regional disparity estimation algorithm via block matching is proposed. Firstly, based on the concept of the depth object, the number of regions is determined and corresponding regional disparities are estimated. Then, the initial image region partition is obtained by minimizing the block matching error. When the number of the blocks in a region is less than a threshold, the region is eliminated, and all the blocks in this region are reassigned to other regions according to the region merging algorithm. The region merging process is iteratively executed to obtain valid region partition results. Experimental results show that the proposed algorithm can effectively partition image regions with different depths and accurately estimate corresponding regional disparities.