In this paper, we propose an image retrieval technique using texture descriptor extracted from a block-based Discrete Cosine Transform (DCT) of the image. To construct the texture descriptor, the absolute DCT coefficients are averaged over the entire image, and the feature vector is formed up by choosing middle-frequency elements of the zigzag-scanned DCT coefficients. This extraction is applied to both query image and all images in the archive. A distance function is used to measure the similarity between two images. This technique is successfully applied to remote sensing images downloaded from www.earthexpolrer.usgs.gov of Thailand regions.
