Research On Color Image Segmentation Algorithm Based On Fuzzy Clustering

In an age when the electronic imaging equipment is extremely convenient, it is urgent for the Image Engineering filed to process a huge number of color images. Image segmentation is one of the most commonly used technologies in image processing, and it is the foundation to ensure the success of image analysis. Literally, image segmentation is a process to divide the pixels and to mark out the region which contains some similar characters in the image. And clustering is the process to divide the objects into a plurality of classification based on some similar properties. The principles between the two are similar, so applying clustering methods to segment image had achieved success. Fuzzy c-means clustering (FCM) algorithm is based on K-means and the fuzzy theory. Because FCM is able to describe the image with high noise, low contrast, and fuzzy spatial relationship, it can be widely used in image segmentation. In this paper, some improvements were made based on FCM, including the introduction of the kernel function, the center initialization collection, and adaptively deciding the number of clustering. By some color image segmentation experiments verifying, this method is effective. This paper mainly aims at the following points:Firstly, the paper briefly illustrates the purpose and significance of the topic. The significance, main research and innovative points of this paper are also explored. By introducing the principle of K-means and FCM, the clustering progress is explained. The second part introduces the definition of image segmentation, color space selection and color image segmentation methods. After having an understanding of the essence of fuzzy clustering and image segmentation, the possibility of applying clustering algorithm to image segmentation was discussed, and the technologies that are needed to achieve adaptively image segmentation are also explored.Secondly, aiming to deal with the disadvantages of FCM, LAKFC, an adaptive kernel fuzzy clustering algorithm based on local research is proposed by the author. This method firstly introduces a nonlinear transformation. The input space is mapped into the high dimension feature space by the kernel function to solve the hardly linear clustering problem. Then using a new center initialization scheme, the approximately iterative hill-climbing method is adopted to optimize the K partitioning clustering. Finally, the evaluation index I(K) based on kernel functions is used to determine the optimal clustering number to realize algorithm self-adaption. The test result of UCI dataset verifies this method effectively.Thirdly, as color image containing a huge number of feature information, it will make the algorithm inefficiency. Aiming to solve the problem, the author adopts SLIC to generate superpixles before segmentation. Image preprocessing contains image enhancement, image restoration, image over segmentation, and etc. Assuming that the image is very clear, the technology to generate superpixels was firstly used to do preprocess, achieving the goal of changing the number of pixels to the number of superpixels. Then, LAKFC was used to cluster and then getting the segmentation result. This method greatly improves the efficiency of image processing, and the result of experiments had verified this statement.