The Multi-granularity Image Clustering Research Based On The Cloud Model

Image segmentation is an important part of machine vision,image processing and other fields.Extracting the target area from the image is the basis of subsequent feature extraction and pattern recognition.The choice of granularity space determines the range of the image information and the accuracy of image segmentation,so there is a big difference in segmentation effect for different levels of granularity.Insufficient and excessive on segmentation is not only the contradiction of traditional segmentation methods,but also the limit of single-granularity segmentation technology.In view of this,the multi-granularity method is utilized to analyze the image information,realizing the automation of image segmentation.However,the existing multi-granularity methods still have transmitted information inaccurately at different levels of granularity.Cloud model,a bidirectional cognitive model,can qualitatively and quantitatively describe the correlation between different granularities.Therefore,the multi granularity clustering model based on cloud model theory is established in this thesis.The model not only improves the accuracy of image segmentation,but also realizes the goal which is to extract the target area adaptively by combining the clustering results of multiplegranularity levels to segment image.The main contributions of this thesis are described as follows:A multi-granularity clustering method based on Gauss transform is proposed.Firstly,Fuzzy C-Means clustering is used to reduce the time complexity of Gaussian cloud transform algorithm.Secondly,the combination of amplitude cloud synthesis algorithm has improved the granularity jump strategy.Finally,the pixels are classified by the improved distance metric.This method solves the problem of multi-granularity and multiscale generation in the remote sensing images.By multi-granularity clustering,the method can adaptively segment remote sensing image more efficiently and more accurately.A multi-granularity clustering method based on region cloud transform is proposed.Firstly,the color image feature is quantified,and its neighborhood information of HSV space is calculated.Then,the improved cloud transform algorithm is utilized to segment remote sensing image from fine-grained.At last,the clustering process from fine-grained segmentation to coarse-grained segmentation is realized by the proposed method.Through this method,the image can be segmented adaptively,and the edge region can be divided into more precisely.