Image retrieval based on weighted texture features using DCT coefficients of JPEG images

The advent of compression standards, such as JPEG and MPEG, has led to the popularity of the compressed form of image data. This has brought on the proliferation of image retrieval techniques in the compressed domain. In this paper, a compressed domain technique for image retrieval, which can extract texture features from DCT coefficients (e.g., within JPEG images) directly, is discussed. It is based on the reordering of the coefficient subbands into a structure similar to the structure of the multi-resolution wavelet subbands. A self-adaptive procedure for generating weights to be associated with different subbands is proposed. A method for assigning suitable weights for new query image is also introduced. Based on weight generation and selection, the performance of texture image retrieval in DCT compressed domain is improved as shown by the experiments carried on the Brodatz texture images.