Image Segmentation Method Based On Cloud Model

Image segmentation is a crucial section in the process of image processing, aiming to extract the specified information from complex image information system. The process of image processing is the process of human cognize image. Due to the uncertainty both on image itself and human-beings' cognition, we cannot conduct precise results based on the existing algorithms in image segmentation.Cloud Model, as a cognitive transformation-bidirectional model between qualitative concepts and quantitative values, reflects the uncertainty of the concept in cognitive processes, and makes fully use of bidirectional cognitive transformation to process it. This study seeks to improve the image segmentation quality on image information and algorithm in terms of Cloud Model, and the details include:(1) Proposing a multi-granularity color method of image segmentation within spatial information-mixing, according to Cloud Model. First, the way of non-uniform quantifying three channels on the HSV to attain one dimensional feature vector and frequency distribution that leads to the multi-particle distribution via reverse-cloud transformation algorithm. Namely, it extracts the initial conception; second, taking region distances and color distances as references for cloud-particle merging through Sobel operator, merges the characteristics of image itself; finally, finishing the target segmentation by the “3En” rules of cloud model. The experimental results show that the proposed method is efficiently to improve the accuracy of image segmentation.(2) Based on Cloud Model, a remote image segmentation method of Gaussian Cloud transform-upgrading is proposed. Remote image has an amount of information and complex structure, which can analyze and operate from multi granularity and multi-layer. The mixed Gauss model converts any kind of frequency distribution functions into superposition of multiple Gauss distribution functions. Cloud transformation which bases on hybrid Gauss is a new method for solving multi granularity problem. Firstly, it uses the K-MEANS clustering method to optimize the selection of the initial granularity; then improving the cloud merging method by amplitude cloud comprehensive method, which means modify the adaptive concept abstraction strategy; finally, separating the granularity of regions, aims to complete the segmentation of remote sensing image. The experimental results verify its correctness and effectiveness.