基于边缘的面向虚拟视绘制的深度编码算法

Edge Based Depth Coding for Virtual View Synthesis

深度图由锐边缘包围纹理的匀质区域组成,因此边缘信息对于虚拟视的绘制至关重要.由于深度图不同区域的失真会对绘制虚拟视质量产生不同的影响,提出一种基于边缘的面向虚拟视绘制的深度编码算法.将深度图划分为边缘区域和平坦区域:对于边缘区域,采用全搜索策略,以提高边缘区域的编码质量;对于平坦区域,采用SKIP模式和16×16模式,以提高编码效率.除此之外,在去块滤波中,提出一种利用纹理图和深度图边缘相似性的可保护深度图片边缘的中值三边滤波器.实验结果表明,与JM编码方案相比,该方法在码率基本不变的前提下,提高了编码效率和虚拟视绘制的主观质量.

Depth maps typically consist of homogeneous areas partitioned by sharp edges representing depth discontinuities, and sharp discontinuities are important in the view rendering. As different depth distortions have different impacts on the virtual view synthesis, an edge based depth coding method is proposed for virtual view synthesis. The depth map is first divided into edge regions and smooth regions. Full search is used for edge regions in which SKIP and 16× 16 modes are used in smooth regions. Besides, a median trilateral filter is designed according to the structure similarity between depth and corresponding video. Experimental results indicate that, compared with JM, the proposed algorithm improves coding efficiency and can achieve high-quality 3D rendering.