Research On A Method Of Graphic Co-segmentation For Distinguishing Noise Images

With the rapid development of the Internet and the widespread use of mobile devices,image data is exploding and single image segmentation can’t meet the needs of large-scale image processing applications.Over the past decade,number of co-segmentaion models and algorithms have been developed to segment multiple image simultaneously and have successfully solved practical problem.However,it makes a slowly process in the problem of multi-object,noise image and complex foreground.Only assuming that image group contains the common object,the traditional co-segmentation can work,and it also have problem at dealing with the complex foreground,similar background and random network image group.So,in this paper,a new co-segmentaion method is proposed,which is optimized by pre-sequence and can recognise noise images.The works of this paper are as follows:Aiming at the problem of traditional co-segmentation be disable to process the noise image.The paper improves a directed graph co-segmentation model based on saliency and the algorithm of the model.By changing the structure of the graph and improve the result path of the graph with the threshold method,noise images can be marked.In the process of constructing the directed graph,similar background problem will be solved by combing color feature with saliency feature.Due to the bad effects of a random sequence of input multi-image,when the graph is constructed,the paper proposed a method that creating a pre-sequence of input multi-image according to the image saliency.To build the pre-sequence,based on the saliency region,the proposed method calculates the similar value of images and then constructs a graph.Computing the average similar value for each image,we select the most similar image as the first image of the pre-sequence.After that,the proposed method gets optimum solution sequence with a greedy algorithm.Being the image sequence,we construct the directed graph.The method can strengthen the relationship between the level of graph,and achieve the goal of optimizating the co-segmentation of noise image distinguished.To test the proposed method,this paper uses evaluation parameters that image retrieval always uses to show the result of our method.The paper refers to three dataset:iCoseg,MSRC and ObjectDiscovery collected by the MIT group.The result of our approach is compared to the several existing methods and it shows our approach is able to effectively distinguish the noise image.