基于图像显著性的立体视频宏块重要性模型  

Stereoscopic Video Macroblock Importance Model Based on Image Saliency

作　　者：陈超[1] 王晓东[1] 姚婷[1] 

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《计算机工程》2016年第1期260-264,共5页Computer Engineering

基　　金：国家自然科学基金资助项目(60832003;61071120);国家科技支撑计划基金资助项目(2012BAH67F01);浙江省自然科学基金资助项目(Y1110161);浙江省教育厅科研计划基金资助项目(Y201327703);浙江省科技厅创新团队自主设计基金资助项目(2012R10009-08);宁波市科技创新团队研究计划基金资助项目(2011B81002)

摘　　要：立体视频中不同区域宏块的重要性不同,部分宏块丢失将严重影响视频重建质量。为此,提出一种结合图像显著性检测的立体视频宏块重要性区分模型。利用图像显著性检测算法计算视频帧的像素显著度,根据运动信息估计视频帧各宏块的重要性,考虑深度信息对宏块重要性的影响,构建立体视频宏块重要性模型。针对不同运动程度和类型的立体视频序列进行丢包仿真实验,结果表明,利用该模型所得宏块重要性指导丢包,解码所得视频的主客观质量相较于随机丢包均有明显提升。The macroblocks of stereoscopic video in different regions have different importance,some macroblock losses will seriously affect the quality of the reconstructed video.For such a problem,a macroblock importance distinguishing model of stereoscopic video is presented,which is combined with image saliency detection.At first,saliency of video frame is calculated by image saliency detection algorithm,then the importance of macroblocks of video frame is estimated according to motion information.Finally,the effect on macroblock importance of depth information is considered,and the macroblock importance model of stereoscopic video is constructed.The simulation of packet loss is conducted on stereoscopic videos of different motion degrees and types.The results demonstrate that,when packet loss is instructed by the macroblock model,the subjective and objective quality of decoded videos are improved compared with random packet loss.

关 键 词：显著性 立体视频 宏块重要性 运动补偿 深度信息 丢包 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]