Research On Saliency Detection Based On Super Pixel Segmentation

With the development of Internet technology,the amount of data increased dramatically,then it is becoming increasingly important to work with computers to process and refine data.How to make the computer simulate human vision and achieve the target has become an urgent problem.Saliency detection is a kind of technology which extracts the main target area quickly and accurately,and eliminates the secondary background area.In recent years,with the development of saliency detection,it has been successfully used in automatic identification,image compression,image segmentation,artificial intelligence and other related fields.The existing saliency detection algorithms have the problem of poor detection in edge segmentation,background culling and target shape extraction.In this thesis,the human saliency attention model,all kinds of image features and prior knowledge are combined to simulate the human vision characteristics.In this thesis,aiming at the above problems,a salient object detection algorithm based on super pixel segmentation and depth feature localization is proposed.The main work of the thesis are as follows:1.Several kinds of super pixel segmentation algorithms are researched.And simulation experiments of these algorithms are carried out.In this thesis,a super pixel segmentation algorithm based linear iterative is identified as a preprocessing algorithm for saliency target detection.2.According to the existing detection methods cannot significantly eliminate background and cannot completely extract target shape,a salient object detection algorithm based on super pixel segmentation and depth feature localization is proposed.In this thesis,color features,intensity features,orientation features,depth features,and the human vision characteristics are combined.Then a location saliency map and a depth saliency map are proposed.Finally,a salient object detection algorithm based on depth feature localization is proposed.Results show that the proposed algorithm is not limited to target characteristics,the environment and other conditions,and the detection effect is more consistent with human significant effect.3.The proposed algorithm,three kinds of algorithms on the classic and five kinds of algorithms on recent years are analyzed by means of Precision index,Recall index and F-Measure index.Results show that the AUC value of the proposed algorithm is 0.6888,which is 7% higher than that of other algorithms.And the detection effect of the proposed algorithm is more consistent with human significant effect.