Autoregressive Estimation of High-Frequency Components in DCT Image Coding

Abstract

As the capacity of communication networks grows larger and the speed of data processing increases, various coding methods for handling digital images conveniently are being developed. The DCT-based standard coding method generally achieves good compression performance. However, for images containing sharp edges, a high compression ratio causes degradation of the reconstructed edges. In this paper, we propose a method for estimating the high-frequency components from the lower frequency components. This method is derived by applying an auto-regressive (AR) estimation to the DCT coefficients of the edge blocks, which are the subblocks containing sharp edges. The AR coefficients are obtained for each edge block. The image is reconstructed from the AR coefficients, the estimation errors of the high-frequency components, and the other frequency components. The AR coefficients are quantized uniformly and incorporated into the head of the AC component sequences in the JPEG standard method. When the proposed method was applied to some images containing sharp edges such as characters or figures, the coding performance was improved, compared with the JPEG baseline coding method.