Genetic Clustering Algorithm And Its Application In Image Segmentation

Image segmentation is to segment an image into some special regions which have no public sector according to gray lever,color or character of geometry,and all the pixels of each region are homogeneous.Generally speaking,to segment an image is very difficult.The current image segmentation algorithm is the technology aiming at the object of segmentation and it is also related to the certain issues.Therefore,scientific researches on image segmentation algorithm become more and more imperative.Because of the complexity and close correlation of the image information,the precision of segmentation is affected at different levels in image processing.In conjunction with genetic clustering algorithm,image segmentation can get a better effect in comparison with the traditional methods.The genetic algorithm(GA) as an intelligent optimization algorithm makes use of the natural selecting and genetic rules for stochastic searching.The primary character of GA is that it can operate at multiple points in the searching space synchronously for the global optimization.The way of searching is led by the stochastic rules and is independent of the gradient information.So GA is capable of solving the nonlinear problems which may cause difficulty for the traditional methods. On the other side,cluster is a kind of unsupervised learning.The goal of clustering is to partition data set into such clusters that objects within a cluster have high similarity and objects in different clusters are dissimilar.Image segmentation is to distil the region interested,so image segmentation can be treated as a kind of clusters.K-means clustering algorithm has powerful local search capability,however it is very sensitive to the initial clustering center and is easy to run into a local minimum.Combining the genetic algorithm and K-means clustering algorithm which just makes full use of the global search capability of genetic algorithm and local search capability of K-means clustering algorithm can find a solution which is easy to be processed,has high segmentation precision and strong practicability.Consequently,this method has great potentialities for further development of image segmentation technology.Based on the application of improved genetic K-means clustering algorithm in image segmentation,the following work has been done in this paper.(1) An algorithm for image segmentation based on the maximum interclass variance of improved genetic algorithm was proposed,aiming to settle such problems as premature or unconvergence by using traditional genetic algorithm.Considering interclass variance as fitness function makes use of the global search capability of genetic algorithm to find the maximum interclass variance to segment an image.First,the algorithm not only save the local optimum in each iteration as candidate of the global optimum,but also let the local optimum in each iteration go to the next iteration without any reasons,which does not replace the unit that has worst fitness.Then,it makes full use of the histogram of the image to be segmentalized as the prior knowledge,thus causing the reduction of the primary population size and the enhancement of seeking priority of the genetic algorithm,which overcome the disadvantage that the traditional genetic algorithm lacks prior knowledge and produces primary population blindly.Finally,the so called "double self-adaptive crossover probability" is used to realize crossover operation,which means making use of crossover probability based on chromosome and gene to make crossover probability self-adaptive changing, and normalize the probability of the crossover in each iteration.So,it not only makes the unity which has high fitness go to the next iteration and the unity which has low fitness wash out in time,and also considers the importance of every gene sufficiently.(2) An improved genetic K-means clustering algorithm is proposed and applied to image segmentation in order to overcome the maximum interclass variance's shortcoming in image segmentation above-mentioned and to realize segmentation of the noisy images.According to the characteristics of the image,the feature vector of the pixel is properly chosen,which takes intensity information and spatial information into account fully,and the feature vector can be extended expediently if it needed.The weight factors of the feature vector are adjusted,which enhances the segmentation precision and practicability.The membership matrix is calculated in a fast way and computing time is reduced after flinging the relationship between membership matrix and the chromosome with binary system coding.The results of the experiments demonstrate that in image segmentation the improved genetic K-means clustering algorithm can be converged to global optimum fast,and it is also immune to image with Gaussian noise to some extent.