ABSTRACT
In order to reconstruct the 3-D shape and camera motion from the 2-D observation matrix, the matrix factorization methods have been widely used in computer vision. In real video clips, these projections however are not always visible along the entire image sequence due to the occlusion. In this paper, we focus on the noise level estimation of the 2-D observation matrix after recovery of the missing data by various recovery algorithms and the evaluation of the accuracy for their recovery algorithms.
