一种基于空时域相关性的HEVC帧间预测模式快速决策算法

为了减小高性能视频编码（high efficiency video coding，HEVC）的复杂度，提出一种基于四叉树结构类型分析的帧间预测模式快速决策改进算法。一方面，统计并分析空时域相邻编码单元（coding unit，CU）和 CU 深度之间的相关性，确定其深度遍历区间，跳过一些冗余 CU 的分割，从而减少 CU 深度划分的复杂度；另一方面，通过分析遍历区间内 CU 的运动特征和预测单元（prediction unit，PU）模式特性，确定不同运动特征所对应的 PU 模式，减少所需遍历候选 PU 模式的数量和所需要进行的率失真代价计算的数量。与 HM14．0相比，采用该算法的 HEVC 编码器在输出比特率增加很小的情况下，平均编码时间降低了46．9％，而编码后的视频质量基本不变。

A fast inter prediction mode decision algorithm for HEVC based on spatio-temporal correlations

In order to reduce the complexity of high efficiency video coding (HEVC),a fast inter prediction improved algorithm based on quad-tree analysis is proposed.On the one hand,we can reduce the complexity of the depth-splitting by utilizing the relationship between spatial-temporal information of CU and the depth of it.And we also can make some redundant CU skipped by ensuring the depth interval.On the other hand,it can make sure the different motion characteristics of the PU mode by analyzing the motion features of CU which is in depth interval and the property of PU mode.So it can reduce the number of ergodic candidates of PU mode and greatly reduce the number of the needed rate distortion cost calculations.Compared to HM14.0,the experiments show that the proposed algorithm of HEVC encoder is able to cut the average complexity of HM14.0 encoder by 46.9% at a small bit rate increase.Besides,the quality of encoded video almost remains unchanged.