Foreground Extraction of Underwater Videos via Sparse and Low-Rank Matrix Decomposition
Hongwei Qin, Yigang Peng, Xiu Li · 2014
In this paper, we propose a new method for foreground extraction of underwater videos based on sparse and low-rank matrix decomposition. By stacking the underwater video frames as columns of a matrix, principal component pursuit algorithm is used for decomposing the matrix into a low-rank matrix representing the stationary background and a sparse matrix representing the activities in the foreground. Then, the sparse matrix is processed with adaptive threshold to extract objects in the foreground. We evaluate our method quantitatively on various underwater videos. Our method is robust to various scenarios like blurred videos, illumination variations in the background, and crowded foreground objects. The experimental results demonstrate the promising performance of our proposed method.