Implementation of Clustering-Based Image Retrieval System

Digital images are useful media for storing spatial, spectral and temporal components of information. Large image databases often store the images in compressed format, JPEG for example. This paper examines the algorithms of direct extraction of low level features from compressed images, working with three different clustering techniques. Results indicate that a K-Harmonic means clustering algorithm not only has the advantage of shorter in both clustering and searching time but also can have a higher accuracy rate. Key-Words: Image compression, DCT domain, clustering, Image retrieval