Research Of Visual Saliency Detection Algorithm Driven By Human Eye Fixation

With the continuous development of Internet technology,the computer needs to handle vast amounts of data,in which 80%of the information is human visual information,so completing the task of image or video analysis fast and accurately is a research focus.This research is driven by human eye fixation and tends to study visual saliency detection models.Human vision mechanism shows that when humans observe the image or video,humans are likely to focus on the area of interest to human eye,while ignore the relatively insignificant area.The main task of visual saliency detection is extracting the areas of interest in image or video as accurately as possible.In general,the detection of visual saliency is divided into two directions,one is salient object detection,the other is predicting human eye fixation.In this paper,research focuses on predicting human eye fixation.Through the study of existing visual saliency detection model,this paper has proposed a video saliency detection algorithm based on random walk with restart.Firstly,we use SLIC superpixels segmentation algorithm to divide each frame of video into superpixels and build an undirected graph.Secondly,we extract the space feature in the YUV color space to establish space transition probability matrix,and extract the motion vector from H.264 stream in the temporal domain to establish the restarting matrix based on the random walk model.Then,we use a boundary prior to establish a restarting matrix based background prior using absorption characteristics of Markov chain.Finally,this research applies improved random walk with restart model to get the final video saliency map.We tested our algorithm in the public SFU and CRCNS video dataset.The experimental results show that our saliency detection algorithm not only has a lower time complexity,but higher detection accuracy and effectively suppress the background noise.Due to the growing concern of visual saliency detection in compressed domain,this paper also proposed a video saliency detection algorithm based on mutual information in compressed domain,the algorithm firstly extracts the spatial feature and temporal feature from the compressed domain.Secondly,we use Shannon mutual information to compute the feature contrast in the center window and the surrounding window for each macro block to get the saliency value.Then we combine the spatial and temporal saliency maps to obtain a spatiotemporal saliency map.Finally,a convex-hull-based center bias is added to optimize the saliency maps.Experimental results show that the proposed method outperforms the existing state of-the-art saliency detection methods.