Video saliency detection using the propagation of image saliency between frames

Abstract

Recently, the saliency detection of image and video has attracted more and more attention due to its wide application. In this paper, we propose an image registration method based on SLIC image segmentation algorithm, which is used to realize the saliency information propagate between frame and frame. First, we use SLIC algorithm to divide each image in video into many regions, obtain the brightness and color information of each frame through LAB color space. Then, use the optical flow method to get the motion information of each frame. At the same time, the saliency information of the first frame is calculated by the method of Li et al [1]. Then, the following procedure is iterated from the first frame: first of all, we perform the registration with the divided regions as processing units, after that, saliency map of this frame is weighted according to the criterion of motion and color and brightness changes, and the result becomes saliency information of the next frame. The salience information obtained by each iteration is generated as a new frame, and the salience result of the video is obtained. The proposed method has been tested in two sets of video sequences and obtained good results.