Research On Image Target Segmentation Methods Based On Clustering

Image segmentation is the most basic and important research topic in image processing, the quality of which directly determines the result of image extraction and recognition. It has been developing very fast and there are already thousands of image segmentation methods available. However, due to the complexity and diversity of the image itself, there is still no common segmentation method for all kinds of images to obtain accurate segmentation results. The image segmentation algorithm based on clustering can analyze the image property to segment the image automatically according to the image feature, the advantages of which makes it obtain wide application in image processing. In this paper, the image segmentation algorithm based on clustering is deeply studied, and the main work and research results are as follows:Firstly, several traditional image segmentation algorithms based on clustering is briefly introduce and simulated, including K-means segmentation algorithm, fuzzy C-means segmentation algorithms and an improved algorithm called Adaptive Fuzzy C-Means Segmentation Algorithm based on the potential function. And the advantages and disadvantages of the three algorithms are analyzed according to the simulation results. In addition, the objective evaluation standard is also presented, providing a strong basis for analyzing the feasibility of the improved algorithms.Secondly, a mean shift segmentation algorithm based on improved FCM is proposed in this paper aiming at the problem that the traditional mean-shift algorithm is easy to fall into local convergence to a over-segmentation result. Firstly, an initial image is obtained by segmenting an original image using mean shift image segmentation method. Then the improved FCM algorithm is used to cluster the initial image and the final segmentation result is obtained. The simulation show that the algorithm could improve the over-segmentation problem effectively and get the desired segmentation, the advantages of high convergence speed and strong anti-noise performance of traditional Mean shift algorithm are remained.Finally, an improved graph theory segmentation algorithm based on region merging is proposed in this paper to solve the problem of over-segmentation using the traditional graph theory segmentation algorithm based on region merging. Firstly, an original image is clustering segmented using region growing method, and then the weighting functions and measure functions in the original algorithm are redefined to implement the secondary segmentation. The experimental results show that the algorithm could obtain the desired segmentation results while ensuring the results meet the global features of the image. And this segmentation method is adapted to targets with complex structures.