Research Of Compressed Domain Based Multi-feature Fusion For Image Segmentation

In recent years,with the rapid development of science and technology,people need to deal with a large number of information(such as images,video,documents,etc.)every day for further analysis and research.As one of the carriers of information,images play an important role in the process of receiving,processing and transmitting information.Since twenty-first Century,with the rapid development of computer technology,artificial intelligence and thinking science,digital image processing technology has been developed towards more rapidly,accurately and deeply direction.Image segmentation is one of the most important problems in image processing and computer vision.It is also one of the most important problems in the field of digital image processing.Image segmentation aims at partition the image into several disjoint regions based on the same color,brightness,texture or other "special significance",each region should satisfy the consistency condition within each area.Image segmentation is one of the key technologies in image processing,and the quality of image segmentation will affect the performance of image analysis directly.Currently,existing image segmentation technology has its limitations such as high complexity and less robustness,too much manual intervention,target segmentation is not accurate with complicated background.Therefore,image segmentation is one of the difficult problems in machine vision technology and the focus of research in recent years.With the rapid development acquisition equipment of digital images,current resolution of digital images is much higher,the size of image is also large;therefore,design a fast and efficient image segmentation method is particularly important.The main content of this paper is the image segmentation method,which is based on graph theory and spectral clustering.The modified discrete cosine transform(Discrete Cosine Transform)is used to preprocess the image,and then the square block DCT(DCTSBS)is obtained,a graph which is corresponding to the image is constructed by using the DCT-SBS as graph node.In order to obtain the multi-feature of each node,the color information,texture and position information of each node is extracted and induced to a novel feature fusion method to calculate weights between nodes efficiently.Finally,the DCT-SBS based clustering algorithm framework is proposed and used to finish the image segmentation task.The main research contents of this paper are as follows:1.It introduces some fundamental approach of graph theory based image segmentation,since traditional segmentation methods usually represent image pixel as graph nodes,with nodes increases the computational complexity greatly increases in exponential way.So in this paper the DCT-SBS is used to construct the graph nodes,construction in this way not only reduce the complexity,but also keep the original structure information of image data at the same time;2.A novel information fusion method based on multi-feature in compressed domain is proposed which can be used to compute the weights between nodes effectively by using the color information,texture information and position information.3.Under the principle of the traditional spectral clustering algorithm,this paper proposed a general framework of the spectral clustering algorithm based on DCT-SBS.It turns out that our algorithm can be transformed into several classic clustering algorithms by redefine the diagonal matrix.4.The proposed algorithm is compared with eight different image segmentation algorithms.In order to validate the effectiveness of the novel information fusion method and our DCT-SBS based spectral clustering algorithm,the Corel1000 database,MSRA10 K database is hired to be segmented by the proposed method and other 8 state-of-the-art segmentation methods.Experimental results show that the proposed image segmentation method has higher segmentation accuracy and efficiency which has certain theoretical significance and application value.