Study On Saliency Region Detection

Saliency measures the quality of the region standing out to its neighbors. Usually, salient regions are those attracting the most attention when observers quickly glance the scene. Without prior knowledge, identifying these regions is a complex problem, and it usually lays in the interdisciplinary of biology, psychology, computer science and other fields. With the effort of predecessors, generated saliency maps have been successfully applied in many multimedia applications, such as image segmentation, object detection, image resizing and Image compression.Firstly, the basic conception of salient region detection and the history of its development are introduced. And the main idea of saliency detection is categorized into several frameworks in order to help to do the further research in this paper.Then we propose a two-stage method for bottom-up saliency detection. In the first stage, we use contrast assumption to find the foreground regions. Image patches are connected in a fully-connect directed graph. A patch ranks high only if it stands out from its neighbors in the image. In the second stage, the saliency results from the first step can be roughly segmented into background and foreground region. We take the labeled foreground nodes as ranking queries. In practice, manifold ranking is used to rank the queries. Based on relevance to queries, the final saliency of each node can be computed.We test our method on one public available benchmark data sets, and the result shows a great improvement comparing to the original contrast-based algorithms. Our method also generates comparable result with the state-of-the- art methods...