Salient Object Detection via Region Shape Feature Contrast and Saliency Fusion

Abstract

The salient object detection has lately received great attention due to their enhancement for many computer vision applications. Shape information plays an important role in the human vision system while it is underutilized in most existing saliency detection methods. In an effort to overcome this challenge, a novel region shape feature descriptor is proposed. As our best known, we novelly model both local and global contrast in one hand-crafted method. What's more, the most saliency approaches may start with an image segmentation method to get the region patches. However the matching degree of the segmented regions and its extracted features has not been argued clearly. The result shows that our region shape feature as a middle semantic feature could represent the region better than color-based method. Weextensively evaluate our algorithm using traditional salient object detection datasets named Oxford Flower Dataset. Ourexperimental results demonstrate that our algorithm improves the performance of state-of-the-art.