Research On Content-based Image Retrieval And Clustering Feedback System

There are two problems to be resolved in content-based image retrieval. One is that the semantic gap between high-level concepts and low-level features has been identified; the other one is that the dimension of feature vector is too high so that it slows down the retrieval speed in huge image database.In the view of the above questions, an image retrieval simulation system which is based on the research of the content-based image retrieval feedback system and feature vector indexing method is designed and realized in this paper. Firstly, a histogram extraction algorithm based on HSV color model is proposed to improve the efficiency of color feature retrieval, the algorithm can decrease complexity dramatically. Secondly, in order to improve the efficiency of retrieving, a new retrieval feedback mechanism based on clustering feedback is proposed, it decided the category of feedback image by dividing the clusters of data set and feedback result, then got more results from the relational category, the method can improve the accuracy of retrieving and increase the retrieval speed in the feedback procedure. At last, a local dimensionality reduction algorithm is proposed to decrease the dimension of high dimensional feature vector, this algorithm decreased the dimension of feature vector in each cluster existing in database respectively by PCA algorithm, and it can decrease the complexity dramatically and increase the efficiency of retrieval.In this paper, the image retrieval system was designed in C# on Visual studio 2005 platform; all of the algorithms were realized in the system, the experiment results showed the efficiency of the new system.