Multi-Level Discrete Cosine Transform for Content-Based Image Retrieval by Support Vector Machines

Texture feature extraction is widely used in content-based image retrieval (CBIR) and is not efficient to be implemented directly in the pixel domain due to high information redundancy and strong correlations in raw images. It is well known that low-frequency coefficients of the discrete cosine transforms (DCTs) preserve the most important image features. In this paper, we use multi-level DCTs (MDCTs) to generate image texture feature vectors for the purpose of CBIR. The texture feature vectors generated from MDCTs coefficients and Zernike moments are classified by support vector machines (SVMs). The experimental result shows good average retrieval accuracy. It also shows that DCT coefficients from low level resolution images are sufficient to extract image texture feature with significant less computing cost.