高效视频编码的快速编码单元深度遍历选择和早期编码单元裁剪

发布时间：2019-03-25 14:28

【摘要】：为了降低新一代高效视频编码(High Efficiency Video Coding,HEVC)标准的编码复杂度,提出了一种基于四叉树结构类型分析和早期编码单元(Coding Unit,CU)裁剪的HEVC快速编码新算法。首先,通过分析已编码的最大编码单元(Largest CU,LCU)四叉树结构类型,确定其深度遍历区间(Depth Range,DR)类型。然后,利用相邻已编码的LCU以及前向参考帧和后向参考帧中坐标位置相同的LCU的DR类型来预测当前LCU的DR类型,并根据预测得到的DR类型对当前LCU设定CU深度遍历区间。最后,采用贝叶斯决策原理获取阈值,并利用该阈值在CU分割过程进行早期CU裁剪。实验结果表明:相对于原始HEVC编码结构,本文算法在随机访问模式下编码时间平均减少41.55%,BDBR(Bjontegaard Delta Bit Rate)只增加约1.94%,BDPSNR(Bjontegaard Delta Peak Signal-to-noise Rate)只降低了0.06dB;与Shen方案相比,该算法可以降低12%左右的计算复杂度,BDBR只增加约1.09%,BDPSNR只降低了0.03dB。
[Abstract]:In order to reduce the coding complexity of the new generation high efficient video coding (High Efficiency Video Coding,HEVC) standard, a new HEVC fast coding algorithm based on quadtree structure type analysis and early coding unit (Coding Unit,CU) clipping is proposed. Firstly, the type of depth traversal interval (Depth Range,DR) is determined by analyzing the quad-tree structure type of encoded maximum coding unit (Largest CU,LCU). Then, the DR type of the current LCU is predicted by using the DR type of the adjacent encoded DR and the DR type of the LCU with the same coordinate position in the forward reference frame and the backward reference frame, and the CU depth traversal interval is set for the current LCU according to the predicted DR type. Finally, the Bayesian decision principle is used to obtain the threshold, and the threshold is used to cut the early CU in the CU segmentation process. The experimental results show that compared with the original HEVC coding structure, the average coding time of the proposed algorithm in random access mode is reduced by 41.55%, and the BDBR (Bjontegaard Delta Bit Rate) is only increased by about 1.94%. BDPSNR (Bjontegaard Delta Peak Signal-to-noise Rate decreased only 0.06dB; Compared with the Shen scheme, the algorithm can reduce the computational complexity by about 12%, increase the BDBR by only about 1.09%, and decrease the BDP SNR by only 0.03dB.
【作者单位】：宁波大学信息科学与工程学院;南京大学计算机软件新技术国家重点实验室;
【基金】：国家自然科学基金资助项目(No.61171163,No.61271270,No.61271021,No.61311140262,No.U1301257)
【分类号】：TN919.81

【参考文献】