Clustering Analysis In Image Classification Of The Applied Research

So far, the classification recognition of the image is still difficulty in the image processing field. Because the real world is diverse and complex, the way of obtaining images is also various; which makes different images vary greatly with each other and it hard to use uniform ways or models describing.The classification of the image used the clustering analysis method in the data mining in this paper. First of all, we have carried out a large number of experimental verification based on Clustering Algorithm of Density and Density reachable (Clustering Algorithm Based on Density and Density reachable, CADD), and improved four deficiencies of CADD algorithm exposed in the application of large-scale image data aggregate. (1) Rewrote the code of the "calculate and save different matrix" to reduce the occupied memory of the preservation of the different matrix. (2) New procedures introduced a new parameter - the number of cluster threshold. (3) Improved the calculating method of the density up to distance, in order to make the original CADD algorithm can effectively process the varying density noise and isolated point. (4) Added a new calculating method of data object similarity metric - cosine similarity methods.Secondly, the experimental results of the improved CADD algorithm and the traditional K-means clustering and Hierarchical Cluster algorithm in true color bitmap BMP classification were compared and analyzed. Come to the conclusion that: (1) The improved CADD algorithm is superior in high clustering precision and resolution compared with the K-means clustering and Hierarchical Cluster algorithm. (2) Despite the CADD algorithm also needs to enter the initial parameters in the clustering process: density parameterσand the initial density up to distance adjustment factor coefR. But the experimental research showed that the changing of density parameterσhad little effect on the results of clustering. By the definition, choosing the initial density up to distance adjustment factor coefR(0