基于纹理信息与深度图空时梯度的深度图上采样算法

摘要

For the 3D video data which are represented as “2D+depth”,down-sampling coding on depth map prior to encoding can reduce the bitrate effectively.However,some down-sampling distortion introducedwill af-fect the quality of the synthesized map.Therefore,effectiveup-sampling of the depth map at the decoder to re-duce the impact of down-sampling distortion is the key issue of the study.In this paper,the depth up-sampling algorithm that jointly employs the correlation between the depth map and its corresponding texture image and the spatial-temporal correlation in depth maps is proposed.The spatial-temporal correlation among depth map pixels is decided based on the time gradient among depth maps,as well as the horizontal gradient and vertical gradient of pixels in depth maps.Sampling reference points are selected to conduct sampling of depth maps. Experimental results show that the proposed scheme improves both depth map coding performance and the quality of final synthesized virtual view image.%针对基于“2D+深度”数据表示格式的3D 视频数据，在编码前对深度图进行降采样编码能够有效地降低码率，然而引入的降采样失真将会影响合成图像的质量。因此，如何有效地在解码端进行上采样以降低降采样失真的影响成为了当前研究的关键问题。本文提出了基于纹理信息与深度图空时梯度的深度图上采样算法，基于深度序列中图像间的时间梯度，图像像素点的水平梯度和垂直梯度，确定深度图像素间的时空相关性，对采样参考点进行决策，从而利用相关的参考点对深度图进行上采样。实验表明，本文方法提高了深度序列的编码性能和最终合成的虚拟视点图像的质量。