面向通用空域错误隐藏的自适应混合填充算法

Universal spatial error concealment via adaptive hybrid filling

在现有的空域错误隐藏算法中,运行快速的多方向插值算法在处理多种丢失模式时效果不甚理想,而恢复质量好的迭代类算法又难以满足实时性需求。为此,本文提出了一种面向通用空域错误隐藏的自适应混合填充(Adaptive Hybrid Filling,AHF)算法,以更好地进行多性能指标的联合优化。对于当前受损块的错误隐藏,AHF算法首先通过各向同性梯度检测器学习延拓区域的邻域梯度特征,根据收缩填充次序执行局部预测过程,从外层素组到内层素组逐一地恢复受损块的各个素组;若预测相关性足够低,AHF算法将转而执行非局部片匹配过程,利用相似片对同样位置的未隐藏像素进行填充。实验结果表明相比于其他代表性的空域错误隐藏算法,AHF算法较好地平衡了通用性、计算复杂度和恢复质量等多指标性能,在典型丢失模式下取得了具有竞争力的综合性能。

Among the existing spatial error concealment algorithms,the fast multi-directional interpolation algorithms are not very effective in handling multiple kinds of loss patterns,while iterative approximation algorithms with good recovery quality can hardly meet the real-time requirements.For universal spatial error concealment,this paper proposes an adaptive hybrid filling(AHF)algorithm to jointly optimize multiple performance metrics.For each missing block,the AHF algorithm firstly learns the local feature information of an extended region through an isotropic gradient detector,and performs a local prediction for each pixel group in a group-by-group shrinkage order.When the prediction correlation is low,the AHF algorithm will switch to a non-local patch matching process and fill the location with a similar patch.The experimental results show that,compared with other spatial error concealment algorithms,the proposed AHF algorithm can well balance the performance in multiple aspects such as versatility,computational complexity,and recovery quality,and thus achieve a competitive performance on typical loss patterns.