Study On Multi-Feature Image Segmentation Based On Mean-Shift And Ncut

Image segmentation is the process of dividing an image into non-overlapping meaningful areas.As the first step of image understanding and pattern recognition,image segmentation is a classic problem in computer vision,and becomes a hot topic in the field of image understanding.So far,there are many image segmentation algorithms.Among them,Mean-Shift algorithm and Ncut algorithm are widely used.Mean-Shift algorithm is an unsupervised clustering image segmentation algorithm with high convergence speed and robustness to noise.However it may cause over segmentation.Ncut algorithm is an algorithm based on graph theory.The Ncut criterion is a global optimization criterion and is mainly applied to pixels to carry out image segmentation.As a result,the computation is large and the Ncut algorithm is not suitable for the real time image processing.Taking the characteristics of the two algorithms into account,we research an image segmentation algorithm based on Mean-Shift and Ncut in this paper.Mean-Shift is used for image processing and then Ncut is applied to regional clustering.The new algorithm can reduce computation and noise interference,which is beneficial to real time image processing.The main research contents are as follows:1.Introduce the principle of Mean-Shift algorithm,including its basic form,expanded form,steps of the algorithm,and discussing the convergence of the algorithm,analyzing the image segmentation based on Mean-Shift algorithm.Finally,simulation results show that the selection of the three parameters Mhh),,(sr in the algorithm has great impact on segmentation results.2.Introduce the relevant basic theory of graph,two kinds of image segmentation criteria(cut and Ncut)and the solution of Ncut.Then this chapter introduces the 2-way Ncut and K-way Ncut algorithm,which base on graph theory.Finally,simulation results show that the selection of parameters K and the weight matrix W in the K-way Ncut algorithm is very important.3.After analyzing Mean-Shift and Ncut algorithm,a segmentation algorithm connecting Mean-Shift and Ncut is proposed to improve the segmentation speed and effects.First,Mean-Shift algorithm is used to preprocess the image.As a result,the image is divided into many over-segmentation regions.Then we use some representative points to replace these regions and construct an undirected weighted graph to describe the relationship of these points.Last,Ncut algorithm is applied to these areas for clustering.Experimental results demonstrate the feasibility and superiority of the algorithm.4.Considering other common features of image,introduce methods of extracting texture features and edge features.Then select special features to take on feature fusion and form multi-feature images.Apply the algorithm connecting Mean-Shift and Ncut to the multi-feature image to get segmentation results.Experimental results demonstrated that the segmentation effects of multi-feature image was better than origin image.