1: \begin{abstract}

2: %Foreground detection in a given video sequence is a pivotal step in many machine learning applications such as video surveillance system. Initiated by the Robust Principle Component Analysis (RPCA), different variants of the low-rank and sparse subspace learning techniques have provided an effective framework for such a task. The focus of this study is to part from the additive assumption of such models and directly recover the mask of the moving object, as in a real-world scenario, the smooth background is overlaid with rather than added to the foreground. We propose the Mask Component Analysis (MCA) algorithm which recovers the desired mask via an optimization formulation and prove its convergence. Additionally, the Robust Mask Component Analysis (RMCA) algorithm for the cases with severe noise corruption is presented. Our study shows significant improvement of the detection with no need for post-processing over the previous frameworks.

3: %\end{abstract}

4: