Research On Content-Based Image Retrieval Using SIFT

Content based image retrieval (CBIR) is one of the fields of Multimedia information retrieval. The difficulty of CBIR is to properly express the contents of the images. Current CBIR systems generally make use of lower-level features like color, texture, shape and space relationship. This paper presents a new approach to extract image features for CBIR, which based on the SIFT features. SIFT features are distinctive local image features. The features are invariant to image scaling, translation, and rotation, and partially invariant to illumination changes and affine. The main points of this paper are as follows:1. Systemically analyzed and concluded the character of SIFT features, innovately apply it into CBIR, and improved the general measurement for image similarity. Experiments have showed that the improved distance measurement is better fit for the image similarity presented in this paper.2. Apply principal components analysis (PCA) to reduce the 128 dimensional vectors to lower dimension for easy data analysis. It is carried out by K-L transform which has the distinction of being the optimal linear transformation for keeping the subspace that has largest variance.3. Referring to the vector approximation based indexing for non-uniform high dimensional data sets approach, the speed of filtering datasets has been accelerated.4. Experimental results show that the robust image retrieval can be achieved by apply SIFT features into CBIR。The potential applications have been generalized simply at the end of the paper.