RGB-D salient object detection via feature fusion and multi-scale enhancement

Abstract

Salient object detection, especially for multi-object detection in complex scene, is a very challenging issue in computer vision. With the emergence and promotion of somatosensory sensors such as Kinect, RGB-D data jointing color and depth information can be obtained easily and inexpensively. This paper focuses on the RGB-D salient object detection. Firstly, the RGB image is converted into Lab color space and superpixels are segmented according to color and merged according to depth. Then, color contrast features and depth contrast features are calculated to construct an effective multi-feature fusion to generate saliency map. Finally, multi-scale enhancement is operated on the saliency map to further improve the detection precision. Experiments on the public data set NYU depth V2 show that the proposed method can effectively detect each salient object in multi-object scenes, and can also highlight the each object entirely.