基于空时域特性的恰可察觉失真模型  

Just Noticeable Difference Based on Spatio-Temporal Characteristics

作　　者：夏光晶 潘晨[1] XIA Guangjing;PAN Chen(Communication Principle and System Integration Lab,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]中国计量大学通信原理与系统综合实验室,浙江杭州310018

出　　处：《电视技术》2018年第12期56-60,65,共6页Video Engineering

摘　　要：[目的]现有像素域JND模型没有充分利用HVS时域感知特性。[方法]基于时域感知特性,本文测量了相对运动、时域持续时间两个视觉注意激励源和背景运动、帧间残差波动强度两个视觉注意不确定源。在自信息和信息熵的统一尺度上,将上述四个异质感知特征参量进行同质化融合,得到时域权重调节系数。[结果]在空域JND阈值的基础上,用时域权重调节系数对其进行调整,得到基于空时域感知特性结合的JND模型。[结论]主观实验表明,本文提出的JND模型优于现有像素域JND模型。The existing pixel-JND models do not fully exploit tlie temporal HVS characteristics.Based on temporal characteristics,the paper measures the stimulus saliency fromtwo stimulus-driven parameters,the relative motion and the temporal duration,and measures the uncertainty from two uncertainty-driven parameters,the background global motion and the intensity fluctuation of inter-frame prediction resiciue.These Our different dimension parameters are jointly fused with high homogeneity on the unified scale of self-information and information entropy to get the temporal weight adjustment factor.Based on spatial JNDthreshold!we get the spatio-temporal JNDmodel by using temporal weight adjustment factor to adjust the spatial JNDthreshold.The experiment results show that the proposed JNDmodel yields significant performance improvement with much higher capability of distortion concealment.

关 键 词：恰可察觉失真(JND) HVS特性 自信息 信息熵 

分 类 号：TN919.81[电子电信—通信与信息系统]