基于深度图压缩的边界自适应上采样方案

在三维视频系统中,为了更好地保护深度图的锐利边界并改善虚拟视图的合成质量,提出一种新的边界自适应性上采样方案来恢复由于在编码前进行下采样操作而丢失的深度信息。通过Sobel算子识别出低分辨率深度图和高分辨率纹理图的水平边界和垂直边界。根据深度图不同区域的特点利用纹理信息来指导高分辨率深度图的重建。实验结果表明本文方法在提高了编码效率的同时也改善了合成视图的视觉效果。

An Edge Adaptive Up-sampling Scheme Based on Depth Map Compression

In 3D video system, in order to better preserve the edge sharpness of depth maps and improve the rendering quality of virtual views, the paper proposed a novel two-step edge adaptive up-sampling scheme to recover the lost information due to downsampling operation prior to encoding. We detected the horizontal and vertical edges of low resolution （LR） depth map and high resolution （HR） texture map by Sobel operation. Then, we utilized the texture information to direct the reconstruction of accurate HR depth map according to the characteristic of different areas of depth map. Experimental results show that the proposed method improves both depth map coding efficiency and perceptual quality of rendered views.