Research Of An Image Retrieval Approach Based On BP Neural Network

This dissertation discusses some key problems on content-based image retrieval. A content-based image retrieval method is presented, emphasizing on feature extraction, image segmentation, and region based similarity measure. The method is proved reasonable and effective by experimental results.On the basis of discussing techniques and state of arts on content-based image retrieval, some problems are studied: We analyze typical image feature extraction techniques in content-based image retrieval, such as color feature, texture feature, shape feature, et al. Retrieval is classification in essence. Splitting the whole process of retrieval into two stages in logic. Efficiency is greatly enhanced by classifying the image in the database before hand. Based on this concept, images are pre-classified into categories before finer retrieval. This problem is solved with BP Neural Network. It is suitable for retrieval task on large database. An image segmentation method based on K-means clustering algorithm is proposed to partition an image into regions, and for similarity measure, a region based Quadratic distance is proposed to solve precision problem.