基于误差预测模型的半自动2D转3D关键帧提取算法  

Key-frame Selection Algorithm Based on Errors Prediction Model for Semi-automatic 2D-to-3D Conversion

作　　者：相凯 吴少群[1] 袁红星[1] XIANG Kai;WU Shao-qun;YUAN Hong-xing(School of Electronic and Information Engineering,Ningbo University of Technology,Nigbo 315211,China)

机构地区：[1]宁波工程学院电子与信息工程学院,浙江宁波315211

出　　处：《软件导刊》2018年第4期82-84,共3页Software Guide

基　　金：浙江省自然科学基金项目(LY16F010014);浙江省教育厅项目(Y201533511);宁波市自然科学基金项目(2017A610109);宁波工程学院"王伟明助创基金"项目(2015005)

摘　　要：关键帧是半自动2D转3D的核心技术,现有方法没有考虑关键帧提取与深度传播间的相互影响,难以最小化深度传播误差。针对该问题,通过光流运动分析,根据颜色差异、运动差异与遮挡误差建立深度传播误差预测模型,提出传播误差最小化的关键帧提取算法。实验结果表明,基于误差预测模型的关键帧提取方法深度传播质量更高,平均PSNR改善了0.6dB以上。Key-frame selection is one of core technologies in semi-automatic 2D-to-3D conversion.Existing methods cannot minimize depth propagation errors since they don′t consider key-frame′s impact on depth propagation.To address this issue,we predict propagation errors based on color dissimilarities,motion dissimilarities and occlusion errors,and select key-frames via minimization of propagation errors.Experiments demonstrate that our method can improve quality of depth propagation via the errors prediction model.The PSNR is improved by more than 0.6 dB compared with existing methods.

关 键 词：3D视频 2D转3D 关键帧 深度传播 误差预测 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]