||As vastly increasing amount of digital images, rapidly declining cost of storage, and explosive growth of the Internet, content-based image retrieval (CBIR) has been intensively studied in the last decades. Though a number of image features based on color, texture, and shape attributes in various domains have been reported in the literature, it is still a rigorous challenge to select a good feature set for image classification. In this thesis, some famous CBIR systems are reviewed, and related issues in the retrieval strategy are addressed. The effective indexing and efficient retrieval are identified as a problem, which serves as the most important criterion in choosing the feature set.
Our work mainly focuses on the use of discrete cosine transform (DCT) as a contribution to fast indexing and retrieval in a CBIR system. We will first show the effective representation of images in DCT domain. Then, to further improve the retrieval speed, a two-stage approach based on DCT is proposed. As the character can be regarded as a gray image, the concept of the two-stage approach is also successfully applied to the recognition of Chinese characters. In addition, a set of weights are used to characterize the relative importance of the features in a query image, which plays an important role in the multiple passes of refining the retrieval. An intensive study of such flexible retrieval, called the fuzzy semantic information retrieval model, is realized in a bird searching system. Finally, the prospects of further work based on the findings of the study are given as a conclusion.