| Image segmentations are foundamental technologies of image processing and computer vision. They are important parts of many image analysis and visual systems and they are also success keys for analysis, understanding and description of images. With the developments of the medical image processing technology, image segmentations in medical applications are more and more important and the segmented medical images have been widely applied in various fields. Based on the specificity of the medical applications and assistance to diagnose diseases correctly, which is relation to the people's lives and health, accurate fast segmentations of medical images have the important theoretical and practical values.This paper studies mind evolutionary algorithm and swarm intelligence, proposes the swarm intelligence based mind evolutionary algorithm (SIMEA) and applies it in image segmentations. The study belongs to the cross-frontier research area of information science, computer science and automation science. This paper mainly includes the following sections.(1) Based on the research of MEA and swarm intelligence, the paper improves MEA by philosophic thinking of swarm intelligence, gives its general process, and proves the convergence of the algorithm in this framework.(2) This paper proposes specific improved strategies for MEA based on the characteristics of swarm intelligence. It introduces the chaos sequence to the initial stages of MEA, gives the improved strategy of chaotic initialization and presents three quantitative evaluation indexes. It proposes group migration strategy, which fully uses the information sharing and enhances the convergence speed. It designs dissimilation strategy with density control, which ensures population deversity and avoids premature. The experimental results of numerical optimization show that the improved algorithm has good performances.(3) This paper proposes image segmentations based on SIMEA. For the problems of Shannon entropy, it gives a new definition of fuzzy exponent entropy, and shows fuzzy exponent image entropy with one or two dimension based on the new definition, which are used as fitness functions during image segmentation optimization. It optimizes parameters of fuzzy membership functions by SIMEA, fulfills image segmentations by membership degrees, reduces the computation time.(4) This paper studies further the application of MEA in medical HRCT image segmentation. During the process of medical image segmentations, it introduces professional medical knowledges by granular computing, segments medical HRCT images by granulating, synthesis and conversion of granules. During the process, it applies SIMEA in granule optimizations to improve the algorithm efficiency.The innovations of this paper are as follows.(1) It proposes mind evolutionary algorithm based on swarm intelligence and gives a general process and the proof of convergence.(2) It presents chaotic initialization strategy and designes three quantitative indexes for initial population evaluation.(3) It proposed group migration strategy and density control strategy, which shares the information fully and achieves the best balance between the global convergence and the searching velocity.(4) It defines fuzzy exponent entropy and image entropy with one or two dimension, which are applied in image segmentation and transform segmentation into parameters optimization.(5) It shows a method of medical HRCT segmentation based on SIMEA and granular computing.
|
PDF Full Text Request