基于T-CNN的3D-HEVC深度图帧内快速编码算法  

Fast intra coding algorithm for 3D-HEVC depth map based on T-CNN

作　　者：于源 贾克斌 YU Yuan;JIA Kebin(Faculty of Information Technology,Beijing University of Technology,Beijing 100124;Beijing Key Laboratory of Computational Intelligence and Intelligent System,Beijing University of Technology,Beijing 100124;Beijing Laboratory of Advanced Information Networks,Beijing 100124)

机构地区：[1]北京工业大学信息学部,北京100124 [2]北京工业大学计算智能与智能系统北京市重点实验室,北京100124 [3]先进信息网络北京实验室,北京100124

出　　处：《高技术通讯》2023年第10期1068-1076,共9页Chinese High Technology Letters

基　　金：北京市自然科学基金(4212001)资助项目。

摘　　要：3D-HEVC标准中引入了具有大面积平坦区域、陡峭边缘和低纹理复杂度特性的深度图。针对深度图编码过程中编码单元(CU)率失真优化导致编码复杂度过高这一问题,本文在分析深度图编码所具有的特点的基础上,构建了深度图划分深度数据集,并提出了一种基于两通道特征传递卷积神经网络(T-CNN)的划分深度预测算法。使用本文提出的算法替换原始编码器中各视点下深度图CU划分模块,可以在一定的率失真性能损失下,将原始HTM-16.0编码器编码时间平均减少76%左右,编码效率得到了显著提升。Depth maps with large flat areas,steep edges,and low texture complexity have been introduced into the 3DHEVC standard.To solve the problem of high encoding complexity caused by coding unit(CU)rate-distortion optimization of the depth map,a depth map partition dataset is constructed by analyzing the characteristics of the coding process of depth map.And a partition depth prediction algorithm is proposed based on the two-channel feature transfer convolutional neural network(T-CNN).The CU division process of the depth map is replaced by the proposed algorithm under each viewpoint in the original encoder,and the encoding time of the original HTM-16.0 encoder is reduced by about 76%on average with certain loss of rate-distortion performance.It shows that the proposed algorithm significantly improves the coding efficiency.

关 键 词：3D-HEVC 深度图 帧内编码 卷积神经网络 

分 类 号：TN919.81[电子电信—通信与信息系统]