Research Of Image Segmentation Algorithm Based On Superpixels Clustering

With the development of human society and the progress of science and technology, people need to deal with a lot of messages from the measured or observed data every day, for further management and use. Image plays an important role in information reception, processing and transmission. It is also the most important and the most effective method to get and communicate the information. Image segmentation is a kind of important technology of processing image information. The input image is segmented into meaningful target areas, so as to meet the needs of people in all situations and applications. Most of the traditional image segmentation method is carried out according to the pixels, which means to describe each pixel information such as color, texture and gradient, but the results of this method tends to be too fragmented. In order to describe the images regional information in a better way, superpixels method is introduced into the image segmentation, which can conform to the human understanding of image visual features better.In this paper, the main work is to use superpixels method for pre-segmentation, then to introduced the affinity propagation clustering and the peak density clustering algorithm for superpixels clustering, and finally to realize the image segmentation. The method not only has good segmentation results, but also effectively solves the clustering algorithm to the problem of large amount of calculation. In this paper, the specific work are summarized as follows:1. In This paper, we firstly introduce the traditional image segmentation method, and analyze the advantages and disadvantages of traditional method. Discuss the significance of superpixels method, summarize relatively popular superpixels segmentation method in currently, and then pointed out that in this paper I choose mean-shift algorithm as a superpixels processing method.2. The affinity propagation clustering algorithm not only has long computing time, high computational complexity, but also cannot be applied in large-scale image processing problems, so I propose the affinity propagation algorithm based on superpixels clustering. Firstly pre-segment the image, and calculate the degree of similarity among the superpixels blocks, so as to constitute a similarity matrix for neighbor propagation algorithm. Due to the use of superpixels block instead of pixels, this method effectively reduces the size of the similarity matrix and greatly improves the speed of the algorithm. Through the experiment in Berkeley’s segmentation database, it shows that this algorithm has obtained the good segmentation effect.3. Computational complexity of the peak density clustering algorithm is too high and difficult to be directly applied to image segmentation, and in order to make up the demerit of the superpixels method as a local algorithm, so I propose the density peaks segmentation algorithm based on superpixels clustering. First step is to pre-segment of images, calculate the values of each superpixel’s decision, establish the peak density clustering decision diagram, and finally cluster and merger the pre-segmented images. In order to solve the problem of algorithm parameter setting, introducing the method which estimates the complexity of the image plane decomposition adjusts the algorithm parameter. Experiments prove that this algorithm has very good segmentation results.