Research Method Of Visual Saliency Detection Based On The Gestalt Rules

With the rapid development of the Internet and the continuous change of hardware terminals,in order to effectively transfer information,a large amount of image data appears in people's lives.But not all image data is helpful to people,so people hope that computers can process these image data as quickly and accurately as human visual systems,and select effective information from the images.Based on the above requirements,the visual saliency detection model has gradually become the research content of many scholars.Visual saliency detection is the initial stage of simulating the human visual system,which can quickly and effectively process image data,quickly extract saliency areas from images,and obtain effective information.Visual saliency detection has a wide range of applications in image processing and other fields,which brings great convenience to image compression and image retrieval.Therefore,it is of great significance to deeply study the visual saliency detection.Many existing visual saliency detection methods do not have the theoretical support of human visual perception and lack physiological basis;many methods only focus on the underlying features and have a single feature fusion mechanism.From these two perspectives,in order to further improve the accuracy of saliency detection and better achieve the detection of complex scenes,this thesis research on the visual saliency detection algorithm,the main research content and results are as follows:1.This thesis proposes a visual saliency detection method based on the regional contrast and contour prior information.First,based on the existing SLIC superpixel segmentation algorithm,the image is segmented at multiple scales;then,the global appearance clues based on the color histogram and the local contrast clues based on the proximity of the regions are extracted separately to fully describe the content of the region significant features;then,for the problem of target confusion caused by small differences in the appearance of regions in complex scenes,a mathematical model is established based on the symmetry and continuity rules of Gestalt theory to describe the salient features of the region contour;finally,by increasing the appearance clue weights of the non-linear fusion mechanism,fusing appearance clues and contour saliency to obtain the final saliency map.Experiments on the MSRA-1000 recognized data set show that this method can effectively improve the target saliency recall and precision,which is better than other algorithms compared.2.This thesis establishes a visual saliency detection method based on Gestalt optimization.First,based on the existing SLIC superpixel segmentation algorithm,the image is segmented at multiple scales;then,based on the color and texture features of the image,the corresponding feature map is obtained and fused to generate a saliency map;finally,based on Gestalt theory proximity and similarity rules establish a mathematical model as a constraint to optimize the saliency region and obtain the final saliency map.Experiments on three recognized data sets,MSRA-1000,SED2 and SOD,show that this method is superior to other visual saliency detection methods in both subjective and quantitative indicators.