Research On The Technology Of Clustering-based Image Retrieval

With the rapid development of network technology and the popularity of digital imaging equipment, large-capacity image database has wide application in various fields, thus promotes the development of image database management. As a result, content-based image retrieval technology (Content-based Image Retrieval, CBIR) comes into being, which purpose is to find the required images from a large image database quickly and effectively. CBIR is to retrieve similar images by deriving image features by computer based on visual characteristics of the image content (including color, texture, shape and spatial relationships, etc.), which involves in computer vision, image processing, image understanding, pattern recognition and many other fields, with an important research significance.We had a summary analysis of the basic principles and framework model about CBIR and its applications fields, and studied its key technologies in-depth, including the color space transmission, the description and extraction of image features (image color feature, texture feature and shape feature),the application of K-means clustering algorithm in CBIR,similarity matching and evaluation standards of retrieval system.In this paper, the major work had been summered as follows:First of all, We proposed a weighed multi-feature of the image retrieval algorithm to describe the image more comprehensively, which sumed the weighted color, texture and shape feature vector. Its aim was to cover by the color,texture,and shape information. The experimental results showed that weighed multi-feature based image retrieval was better than the single feature of the image retrieval, which improved the precision and recall of the system.The second,in the design and implementation of the system,We increased a clusering module,by improving the traditional k-means clustering algorithm and then made the improved clustering algorithm into the content-based image retrieval system. By clustering the image in the image database according to some similarity criteria, the similar images were gathered to a class, and different images were classified into different categories, which can not only narrow the search range greatly by filtering out irrelevant images, but also promote the accuracy of image retrieval. The experimental results show that the clustering-based image system performance superior to the traditional retrieval systems,Simulation results also proved the efficiency of algorithm in the image retrieval system. By using Visual Studio2010as a development tool, we designed and realized a weighted multiple-features based image retrieval system in VC++, at the same time, validated all of the image retrieval algorithm involved in the text.