Convex Decomposition Model for Salient Object Detection

An efficient background-salient object separation using background priors is proposed in this paper. When the image is cluttered, it is difficult to decompose the image into foreground and background regions. In order to overcome this limitation, feature vectors are computed from the image using Simple Linear Iterative Clustering algorithm (SLIC). In the next step feature matrix is decomposed into low-rank and sparse component using a convex model. The separation between background and salient object is increased by incorporating background priors into the model. The background priors chosen are location, color and boundary connectivity. In contrast with the bench mark methods, the proposed method delivers accurate result at a faster rate.