A BRMF-based model for missing-data estimation of image sequence

详细信息    查看全文

• 作者：Yang Liu^aAuthor Vitae ; Zhan-Li Sun^a ; ^{zhlsun2006@126.com}Author Vitae ; Xia Chen^aAuthor Vitae ; Li Shang^bAuthor Vitae
• 关键词：Missing-data estimation ; Structure from motion ; Matrix factorization ; Weaker estimator
• 刊名：Neurocomputing
• 出版年：2017
• 出版时间：8 March 2017
• 年：2017
• 卷：228
• 期：Complete
• 页码：65-70
• 全文大小：827 K
• 卷排序：228

文摘

How to effectively deal with occlusion is an important step of structure from motion. In this paper, an accurate missing data estimation method is proposed by combining Bayesian robust matrix factorization (BRMF) and particle swarm optimization (PSO). As BRMF is primely designed for outlier detection, in the proposed method, the missing entries of the observation matrix are firstly replaced by the values that are significantly larger than the non-missing entries. Then, the low-rank factorization matrices are computed via the BRMF to recover the observations. One issue of the BRMF model is that the fluctuation of output results caused by the variation of rank values. Analogously to the classifier-committee learning algorithm, a BRMF-based weaker estimator is constructed to alleviate the unfavorable condition. Moreover, a PSO-based weighting strategy is devised to integrate the outputs of weaker estimators. Experimental results on several widely used image sequences demonstrate the effectiveness and feasibility of the proposed algorithm.