Research On Visual Saliency Detection Model

With the rapid development of intelligent terminals and Internet technologies,the data volume of multimedia resources such as images and videos have increased swiftly.Faced with these large amounts of image data,how to let the computer understand the content of the image scene,smart batchly processing,rapid retrieval with consult and classification with standard target that have became a hot research field in computer vision.The most basic and important step in making the computer intelligently process image which is the detection of the saliency target of the image.This application can promote the follow-up research of the target such as identification,classification,and tracking.The purpose of the visual saliency detection is how to quickly locate the region where the human visual system was interested in the image,so that the content of the image data can be effectively reduced,and in the subsequent processing,the region will subject to more complex processing.Based on the cognitive psychology and neurobiology theory of the visual attention mechanism of the human visual system,this paper proposes an algorithm which based on conditional random field to fusion background priori information to improve the accuracy and robustness of saliency detection.The main work and achievements are as follows:(1)Analyze the background and status of image saliency detection research.From the perspective of cognitive psychology and neurobiology,the concept of visual attention mechanism,visual saliency and other concepts are introduced.From the perspective of the image,introduced basic concepts such as color,texture,shape features and superpixel segmentation algorithms,and the innovation of the classic saliency detection algorithm was conducted briefly.(2)The background information is essential feature distinguish the foreground from images,especially when the image contains multiple targets or complex backgrounds.In this paper,we used the superpixel segmentation technology to segment the imagento a number of uniformly shaped local superpixel regions efficiently.We proposed a novel saliency detection via fusing a set of background features.Our algorithm firstly extracts a set of background features from the input images,which include background uniqueness information,dense information,and sparse information.Finally,using the fusion framework of conditional random fields,three background priori features are taken as input values,and the final saliency map is obtained after data fusion.We evaluate our method on the proposed datasets,and the experimental results show that our approach well against the previous methods in terms of precision and recall.