Salient Object Detection via Graph-Based Flexible Manifold Ranking

Abstract

The task of saliency detection is to segment salient objects in natural scenes. Simple and effective saliency detection model has always been a challenging problem. We explore a graph-based flexible manifold ranking approach for single image saliency detection. An input image is represented as an undirected graph. Feature vectors are extracted covering regional color and texture. An optimal function is used to infer the labels based on linear classification projection and manifold ranking in our work. The optimal function further ensures the reliability of the prediction results. Extensive experiments on four benchmark datasets show that our method is better than the other eight classic methods. So the proposed method is a competitive method.