Research On Techniques For Visual Saliency Detection Based On Gestalt Theory

Saliency detection can be used to reduce the storage resources and improves the utilization of digital media resources by detecting key information in multimedia content and filtering out the redundant information.Currently,researchers have proposed a large number of effective image saliency detection models,but there is little research on video saliency detection.Compared with images,video has an additional dimension of complicated temporal information.How to effectively extract spatiotemporal features in video and fuse them to construct an effective video saliency detection model is still a challenge in computer vision.In addition,with the rapid development of multimedia technology,the research of 360-degree images has attracted extensive attention of researchers.For this particular type of image,how to construct an effective 360-degree image saliency detection model is of great significance for immersive multimedia applications.For video and 360-degree image,two new saliency detection models are proposed respectively.The main contributions of the paper are as follows.(1)This paper proposes a video saliency detection model based on Gestalt theory.The spatial saliency map and the temporal saliency map are calculated by extracting low-level visual features,and then the spatial saliency map and the temporal saliency map are dynamically adaptively integrated based on the similarity law and the common fate law of the Gestalt theory.Compared with the traditional linear fusion method of spatial and temporal saliency map,this method can detect the salient regions in the video more accurately.The experimental results on three public datasets show that the performance of the proposed method is superior to other state-of-art algorithms.(2)This paper proposes a 360-degree image saliency detection model based on Gestalt theory.First,the image is segmented into superpixel blocks by using superpixel segmentation algorithm;the superpixel-level feature contrast is then calculated based on the foreground-background law of the Gestalt theory;the concept of 360-degree boundary connectivity is defined as background measurement.The final saliency map of 360-degree image is predicted by combining feature contrast and boundary connectivity.Experiments on the public 360-degree image databaseshow that the performance of this algorithm is superior to other state-of-art algorithms.