Low Complexity Image Matching in the Compressed Domain by using the DCT-phase

This paper presents a new, simple, and efficient algorithm for image matching in the compressed domain. The proposed technique operates directly on transformed frequency-domain data, which makes it a good choice when dealing with standard-compliant Discrete Cosine Transform (DCT)-based compressed images since it avoids the large computational overhead associated with the decoding stage. Furthermore, the algorithm itself features a very low computational complexity and a high matching accuracy. An example of the effectiveness of this algorithm is illustrated in the frame of a particular content-based image retrieval application. The proposed method exploits the fact that the phase of the coefficients of a DCT transformed image contains a significant amount of information of the original image. By processing only the phase part of the DCT coefficients, a simple and yet high matching accuracy method is achieved. The underlying principle can also be applied in other computer vision applications and in video processing in the compressed domain.