Improved salient objects detection based on salient points

In this paper, we propose an effective framework for detecting salient regions of an image via salient points and color contrast. First, the boosting Harris is used to detect highlight points as the salient points, and the convex hull is constructed to separate the foreground regions and the background regions. Due to the fact that superpixel is perceptually more meaningful than pixel, and which can reduce the complexity of image processing. So we utilize the superpixel algorithm to reprocess original images, and compute the contrast between superpixels in the foreground regions and the background regions to get the salient map. After that, the fusion is used to improve the detection results. Finally, the proposed method is evaluated on the MRSA(Microsoft Research Asia) dataset, and can produce promising results compared with 7 state-of-the-art salient object detection approaches.