Web Image Mining Of K-NN Classification Algorithm Using Relevance Feedback

Data mining (DM) extracts knowledge from a number of data. Along with internet popularization, broad band using , mobile communication development and a amount of multimedia appearing, It becomes a new subject how to mine multimedia data including similarity Search in Multimedia Data, dimension analysis, classification analysis, prediction analysis as well as association analysis of multimedia data.With world wide web development, a great deal of image appear on it every day. Utilizing sufficiently need supporting of image search technique. Research of image search technique began in 1970s. It depends mainly on text method that using key words express image content firstly, then it searches image based on DBMS. This method meets need of image search at a certain extent, but there are two flaw: 1 key words embody subjective feeling because different people have different feeling according to the same image. 2 The task of manual labeling image is very tedious.Therefore, researchers proposed content-based image retrieval (CBIR) that means not to label image by manual way but to create index of image according to image content such as color, shape, texture and so on.Content-based image retrieval is a research hotspot recently. Image mining apply classification of data mining , which improve the efficiency of CBIR. K-Nearest Neighbour is a typical classification method.Visual feature is broadly applied in CBIR, which is only used for low-level feature such as color , shape , texture and so on. there is a gap between low-level feature and high-level semantic knowledge. Though sometimes finding connection between both in special fields , e. g. face and fingerprint, people can't find connection mostly. The terminal user of CBIR system is human being, thus it is important to capture image knowledge from the point view of psychics. Relevence feedback is used for CBIR in order to embed user model into image search.In this paper author applies a algorithm:relevance feedback k-nearest neighbour algorithm, utilizing image from www. google. com by key word image search. Moreover , author implements a simulation system. Experiment result show that author's algorithm can improve precision of image mining.