The technique of content-based image retrieval (CBIR) was come into being with the steady growth of multimedia technique, the main content of this technique is to retrieval relevant images from image database based on automatically derived image features. In recent years, with the development and spread of many compression standard (JPEG, MPEG, H.261, etc), compressed image was used more and more popular and widely. So retrieval operation directly in compressed formatted image becomes a new important trend of CBIR.In this dissertation, lots of exploratory research work has been done around some key techniques of Image Retrieval Based On Compressed-Domain, which include based on Discrete Cosine Transform (DCT) , based on Vector Quantization (VQ) combine with low-level feature extraction, similarity measure and so on. The emphasis of this dissertation:Firstly, some feature extraction algorithms based on color and texture are analyzed and discussed, and made a full-scale discussion of the current compressed domain retrieval techniques.Next, In technique of image retrieval based on DCT compressed-domain. An image retrieval approach based on DCT coefficients reorder is proposed.Finally, In technique of image retrieval based on Vector Quantization, analyzed and compared scalar quantization, vector quantization and classified vector quantization using statistical features for the performances of image retrieval.The presented study is the current research hotspot of image retrieval. Thus its research has both theory and application value.The contribution of this dissertation:An image retrieval approach based on DCT compressed domain is proposed. First, reorder DCT coefficients using multiresolution wavelet transform, then build subband energy histograms formed from reordered DCT coefficients of database images, build indices of images by using Morton order and order database for indexing by using variant B-tree data structure. Many experimental results show that this approach is fast and effective. |