基于分组鲁棒主成分分析的老电影修复  

Group-based vintage film inpainting using robust principal component analysis

作　　者：于冰[1,2] 丁友东[1,2] 董荪[1,2] 黄曦[1,2] 

机构地区：[1]上海大学上海电影学院,上海200072 [2]上海大学上海电影特效工程技术研究中心,上海200072

出　　处：《上海大学学报（自然科学版）》2017年第3期315-323,共9页Journal of Shanghai University:Natural Science Edition

基　　金：国家自然科学基金资助项目(61402278);上海市自然科学基金资助项目(14ZR1415800);上海大学电影学高峰学科和上海电影特效工程技术研究中心资助项目(16dz2251300);上海市科委科技攻关资助项目(16511101302)

摘　　要：以老电影视频为研究对象,针对序列中存在的多种损伤类别,提出一种基于分组鲁棒主成分分析(robust principal component analysis,RPCA)的统一修复方法.采用镜头分割和去闪烁实现对视频序列的预处理.在多分辨率金字塔框架下,采用时空域分组的方式在最粗糙层构造观测矩阵,依次执行基于交替线性法的RPCA变换后,根据帧间误差信息得到大面积破损位置;利用上采样方式构造初步修复结果序列、破损掩模序列以及最近邻偏移矩阵集合,继而对原始序列进行修改,重复时空域分组RPCA变换,实现对老电影视频序列的修复.实验结果证明,该方法能够同时修复画面中的不同损伤,并取得良好的效果.In this paper, a new group-based method using robust principal component analysis （RPCA） is proposed to deal with multiple categories of damage in vintage film sequences. Pre-processing of the video sequence is achieved by shot segmentation and flicker elimination. In a framework of multi-resolution pyramid, an observation matrix is constructed on the coarsest level by space-time domain grouping. After performing RPCA transform based on the alternating linear method in sequence, locations of large area damage are obtained based on inter-frame error information. An initial inpainting result sequence, a break mask sequence, and a nearest neighbor offset matrix set using an upsampling method are constructed. The original sequence is then modified. By repeating the space-time grouping RPCA transform, inpainting of the vintage film sequence is realized. Experimental results show that the method can simultaneously repair different damages in the screen with good performance.

关 键 词：老电影 视频修复 鲁棒主成分分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]