An iterative approach to recovering the missing data in a large low-rank: Application to SFM
Date
2004
Authors
Chen, Pei
Suter, David
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Report
Citation
Statement of Responsibility
Pei Chen and David Suter
Conference Name
Abstract
In the field of computer vision, it is common to require operations on matrices with “missing data”, for example because of occlusion or tracking failures. In this paper, we consider a special case, where the large matrix should be of low rank if it is noise free. This constraint often exists, such as in the factorization method for the problem of structure from motion (SFM). In this paper, we propose a new iterative solution method to the missing-data problem. It has the following advantage: (i) Fast convergence. (ii) The recoverability of the unknown entries can be easily determined. (iii) The initial result, after the initialization step, is exactly correct and no iteration step is required if the data available is noise free and the incomplete matrix is recoverable. We compare the performance of the proposed method with Jacobs’ method. The iterative algorithm performs much better than Jacobs’ method when applied to both synthetic data and real data. Moreover, even after merely the initialization step, the proposed method usually exhibits a better performance than Jacobs’ method
School/Discipline
School of Computer Science