Salient object detection based on an efficient End-to-End Saliency Regression Network

Abstract

Abstract Salient object detection aims at detecting and segmenting the most salient objects from images or videos. It serves as a pre-processing step for a variety of computer vision and image processing tasks. Therefore, efficient and simple detection procedure is the primary requirement of salient object detection. Although many methods with impressive performances have been proposed, they always include complicated procedures. They are time-consuming and not easy to be applied in practical application. In order to address this issue, we propose an efficient and simple salient object detection architecture based on saliency regression network. Our method is a simplified end-to-end deep neural network without any pre-processing and post-processing. It can directly predict a dense full-resolution saliency map for a given image with a compact pipeline. Experimental results on five benchmark datasets show that the proposed method can achieve comparable or better precision performance than the state-of-the-art methods while get an improvement in the detection speed.