Research And Realization Of Image Segmentation On Color Topographic Maps Based On Cluster And Region Grow

Color segmentation technology application domain involves all aspects of human life,include computer vision, physiology, cognitive science and computer science, and other disciplines,which is a combination of theory and application of image processing and recognition technology. And the color of color map image segmentation of target detection based on geographic information system to provide the unique valuable information, along with the development of computer technology, electronic map gradually replace paper maps bring convenience for human life, get the electronic map and color segmentation of color map is the key technology, but also the precondition of vector quantization steps of color map. Theory and practice for many years for the further research of color map image segmentation has laid a good foundation, however,due to its highly mixed color color map image and fuzzy characteristics of vector quantization of color segmentation and further to it brings many problems.This thesis first analyzes significance of color map image segmentation research and current situation of the development of both at home and abroad, points out the main problems, and further study of the current main color map image segmentation algorithm, clustering algorithm and seed region grow algorithm, based on that, this thesis proposes a new method of color segmentation based on GK clustering and seed region grow algorithm, in order to realize the color map image segmentation more accurately. Specific ideas are as follows: first, the image sampling data set to obtain a sample; Reuse the GK clustering algorithm to obtain more accurate clustering center.Secondly using GK clustering algorithm to get the clustering center and on the basis of similarity and spatial connectivity of pixels to define initial seed set.Then according to the growth of certain rules for seed region grow, finally according to the result of the seed region growing to get with the same color information and spatial connectivity layer.GK fuzzy clustering based on covariance matrix of the adaptive distance measurement,can accurately simulate the sample clustering of the spatial distribution of super ellipsoid.This thesis puts forward the improved GK clustering algorithm, through theellipsoid fitting original data sets, with the fitting of covariance matrix instead of the original GK clustering of covariance matrix, to effectively avoid the repeated computation in the clustering algorithm of covariance matrix, shorten the operation time.Based on improved seeds of GK clustering and region growing image segmentation algorithm does not need according to the experience of prior parameters, and the image sampling and ellipsoid fitting methods of covariance matrix effectively shortens the time of clustering and eventually be able to realize the map image automatic segmentation, get map image, and has high accuracy and robustness of universal applicability. In addition, this article proposed algorithm and Stefan Leyk and Ruedi Boesch region growing algorithm is put forward by the comparison and analysis, the results show that the proposed a new image segmentation algorithm has higher accuracy and adaptability. The author has been transplanted the algorithm into xi ’an institute of a Map GIS K9 geographic information system platform, and applied to the automatic image segmentation of color map.However, the clustering algorithm inevitably has the existence of the complexity of algorithm.The study that how to improve the accuracy of the clustering algorithm and shorten the time complexity and better initial seed set is one of the authors emphasis of further research.