Research On The Algorithm And Application Of Finite Hybrid Model Based On Intelligent Optimization

Model based clustering analysis has important guiding significance for pattern recognition,data mining and machine learning,whereas finite mixture model(FMM)plays an important role in image segmentation as the basis of clustering analysis.Aiming at the common problems of image segmentation and the characteristics of the finite mixed model,in this paper,we focus on the image segmentation algorithm based on the asymmetric finite mixture model by breaking through the key technologies.The following researches are carried out:(1)By comparing the Gaussian mixture model with the student's t-distribution model and the generalized Gaussian mixture model,the advantages and disadvantages of these distribution functions are discussed.The solution of the model is optimized by using different intelligent optimization algorithms.The intelligent optimization algorithms such as Genetic algorithm,particle swarm optimization and moth flame(MFO)algorithm are compared in the Gaussian model.We point out the limitations of traditional EM algorithm for solving FMM problem based on the literature research and theoretical deduction.It is explained and proved that the intelligent optimization algorithms based on the population can obtain the optimal solution of the optimized function.(2)We proposed a label transfer and MFO based Gaussian mixture model for image segmentation.On the basis of the MFO optimization for the Gaussian mixture model,the label transfer of the affinity map is used to introduce the global spatial information and solve the noise problem of the image.Next,the bias field estimation algorithm is introduced.We combine the parameters of the Gaussian mixture model with the bias field algorithm to solve the problem of the intensity inhomogeneity.The proposed algorithm is applied to the geographic atrophy segmentation based on SD-OCT image.The advantages of the proposed algorithm are analyzed by compare the segmentation accuracy and other indicators with the existing algorithm.(3)We proposed a rough set bounded asymmetric Gaussian mixture model with spatial constraint for image segmentation.Based on the rough set theory,a new bounded indicator function is proposed to determine the bounded support regions of the observed data.The bounded indicator and posterior probability of a pixel that belongs to each sub-region are estimated based on the rough regions.The within-and between-cluster spatial constraints are introduced by incorporating the spatial information with adaptively selected direction in order to reduce over-smoothness for segmentations.The models are applied to synthetic images,human brain MR images and color images,respectively.Comparsion results by quantitative analysis prove the superior performances of the proposed model.