Content-based Image Retrieval Methodology And Implementation

As the development of information society, the utilization of images has permeated all walks of life. The ever growing source of images has provided people with redundant information. The requirement of rapidly searching for useful images is becoming more and more urgent. The current popular network search engines are mostly based on characters, and the research of image based search is still in its infancy. The implementation of combining digit image processing, database and image retrieval technology together to develop fast, convenient image search engines is of great theoretical and practical values.Content based image retrieval technology includes several aspects:image bottom feature extraction, similarity matching, and index mechanism. This thesis has commenced the study on image bottom feature extraction, which mainly concerns improving and merging the classical algorithm of color feature extraction, texture feature, and shape feature. First, Haar small wave was used to execute a first level decomposition of images, after which a new texture extraction method, which describes textures by calculating its histogram, was developed based on the traditional method that describes texture feature by calculating the energy of high frequency images after small wave decomposition finished; the low frequency images were then partitioned by utilizing Watershed algorithm, which was to separate objects out of the background. Based on this, color features were extracted using key color method, avoiding the traditional method drawbacks of mandatory division; finally, the shape features were extracted by using chain code method. By utilizing feature extraction algorithm mentioned above, great convenience could be provided in extracting various features of images, which dramatically increases the image retrieval accuracy.The present work has used Visual Studio 2005 as a development tool, and oraclelOg as database, to develop a simple content based image retrieval system. The designed CBIR system includes elements not only the feature extraction of images which is based on the above algorithm, but also a personalized search method that allows searches based on parameters inputted by customers. A set of experiments have been carried out on the system, and some good effects are obtained.