3D-HEVC深度图帧内预测快速编码算法

三维高效视频编码在产生了高效的编码效率的同时也是以大量的计算复杂性作为代价的。因此为了降低计算的复杂度,本文提出了一种基于深度学习网络的边缘检测的3D-HEVC深度图帧内预测快速算法。算法中首先使用整体嵌套边缘检测网络对深度图进行边缘检测,而后使用最大类间方差法将得到的概率边缘图进行二值化处理,得到显著性的边缘区域。最后针对处于不同区域的不同尺寸的预测单元,设计了不同的优化方法,通过跳过深度建模模式和其他某些不必要的模式来降低深度图帧内预测的模式选择的复杂度,最终达到减少深度图的编码复杂度的目的。经过实验仿真的验证,本文提出的算法与原始的编码器算法相比,平均总编码时间可减少35%左右,且深度图编码时间平均大约可减少42%,而合成视点的平均比特率仅增加了0.11%。即本文算法在可忽略的质量损失下,达到降低编码时间的目的。

Fast intra coding algorithm for depth map in 3D-HEVC

3D high efficient video coding(3D-HEVC)produces high video coding efficiency but at the cost of heavy computational complexity.Therefore,to reduce the computational complexity,a fast algorithm for 3D-HEVC depth maps intra prediction based on edge detection of deep leaning network is proposed.Firstly,we use the holistically nested edge detection(HED)network to perform edge detection on the depth maps to find edge locations.Secondly,the edge probability map of the output of the HED is binarized using Otsu method to obtain a significant edge region.Finally,different optimization methods are designed for different sizes of prediction unit indifferent regions.By skipping the depth modeling mode and some unnecessary modes,the complexity of the mode selection of the intra prediction is reduced,thereby reduce the coding complexity of the depth maps.Experimental results show that compared with the original encoder,the proposed algorithm can reduce the encoding time by 35%on average,and the encoding time of the depth video by 42%on average,while the average bit rate of the synthesized views is only increased by 0.11%.That is to say,the proposed algorithm achieves the purpose of reducing the coding computational complexity under negligible quality loss.