A still image coding method in which AR model estimation is applied to dct coefficients of edge blocks

Abstract

Image coding technology has been developed for various purposes with the progress of large-scale communication networks and recording technology. Notably, a method using a discrete cosine transform (DCT) and a variable-length code (VLC) has been popular for still image coding. Although this method is excellent for general purposes, some images such as the edges of letters become blurred unless the code compression rate is sufficiently high. This paper proposes a method of estimating high-frequency components from low-frequency components and auto-regressive (AR) coefficients by applying an AR model to the DCT coefficient of an image block (or “block edge”) containing sharp edges. The method produces AR coefficients for each block and preserves the AR coefficients, DCT coefficients and the estimated errors of high-frequency components as data to reconstruct images. The newly introduced AR coefficients are quantized uniformly and added to the head of the components in the standard method. The estimated errors are added to the data instead of the high-frequency components. Examples of images containing sharp edges such as letters were processed by the proposed method, and the results confirm the usefulness of the method.