Research And Application Of Clustering Algorithm In Image Indexing

With the high-speed development of network technology and digital multimedia technology, and the rapid growth of various types of information, people live in a world around a large number of images, resulting in more and more processing requirements of image retrieval. However, traditional text-based image retrieval has not adapted to the current environment. Content-based image retrieval (CBIR) came into being. CBIR extracts the high-dimensional image features automatically, and then does approximate image matching to find the target image. It is a kind of image retrieval technology which integrates a variety of other technology.First, this paper introduces the background, development and current research of CBIR, and summarizes a common content-based image retrieval framework. And then this paper sets forth two core issues: feature extraction and representation, similarity measure.In the image database, images are indexed with high-dimensional feature vectors. Many researches have study plenty of tree-like indexing structures so that it can meet the request of effective image retrieval. Building an indexing structure in image information retrieval becomes a very challenging issue.This paper introduces kinds of clustering methods, the concept of wavelet analysis and some typical high-dimensional index structures. After all, from the combination of a clustering method and wavelet transformation, a new index structure called CBB-tree is proposed. It is an improvement of cluster tree which takes advantage of multi-scale searching performance combined with the good I/O performance of B+ tree.On the basis of this new CBB-tree, the author designs and realizes a content-based image retrieval experimental system. A large number of comparative analysis to experimental results prove that the proposed indexing structure can improve the efficiency of image retrieval.