Improved Hierarchical Clustering Method And Immune Genetic Algorithms-based Self-adaptive Image Segmentation

As a great optimizing method for overall searching, genetic algorithm has beenwidely adopted. In actual application, however, it needs to be improved because itcannot guarantee the overall optimal solution; the best individual in a group is likelyto be dropped and early convergence tend to happen in the developing process. In thefield of image segmentation, genetic algorithm is usually used to establishthresholding segmentation. For the purpose of solving the problems mentioned aboveeffectively, we take the genetic algorithms and the immune system together into ourconsideration. That is, we introduce the various characteristics of immune system intothe framework of genetic algorithms so as to get the most effective optimal solution. This paper studies a new self-adaptive image segmentation method of geneticalgorithms through the improved hierarchical clustering method and artificial immunesystem in the field of image segmentation. On the basis of traditional segmentingmethod, our study realizes a learning technology enlightened by biological immunesystems through the imitation of the natural defensive mechanism. This method offersthe evolution learning mechanisms such as noise endurance, non-teacher learning,self-organization, self-memorization. Also it conquers the disadvantages in traditionalmethod such as the inability of accommodating individual varieties and guaranteeingthe convergence of probability. The main work and result in this paper are described as the following: Aiming at the defects like slow-speed and low-efficiency of the traditionalhierarchical clustering method in pattern-recognition, this paper provides an improvedclustering method by the quartation in the particular image-processing field. Furthermore, this paper offers us a new image-segmenting method on the basisof immune genetic algorithms. According to the regional consistency that the greylevel of the image is distributing, we first make clusters for the images in theirappointed threshold by our improved hierarchical clustering method inpattern-recognition, and then we make the self-adaptive thresholding segmentationthrough integrating the genetic algorithms and making use of the inside functions ofself-accommodation, antigen-recognition and memorization within the immunesystem. Finally, we adjust the threshold value to get the optimal result by imposingsufficiently the advantages of overall searching in genetic algorithms. The realization of the programs in this study is exploited on the environment ofVC++ and NET. The image for experiment is the classic 256*256 grey bitmap in thefield of image-processing.