The blocking effect, which results from independent coding of each subimage in the transform coding, becomes highly visible when the bit rate is sufficiently low. The blocking effects often cause the dominant degradation of the reconstructed image. In this paper, an efficient transform coding technique is proposed by exploiting interblock correlation. Since smooth area of the image is composed of highly correlated blocks, some lower 8?8 DCT coefficients of these blocks have correlation each other. In the measurement of correlation coefficient distribution about lower DCT coefficients, it is noted that reasonable interblock correlations exist in DC and lowest frequency AC coefficients, To exploit such interblock correlation, second DCT was performed for DC, lowest frequency AC coefficients, respectively. Two methods are proposed according to segmentation method of the superblock which is ensemble of the correlated blocks. One uses a superblock of fixed size and variable number of interblock second DCT's using the DCT coefficients, and the other uses variable superblock size according to smoothness of the local characteristics of the image. The performance of two proposed methods are compared with other existing methods (LOT, AC prediction method of JPEG). Since the proposed methods utilize the interblock correlation in the transformed domain efficiently, they improve visual quality of the reconstructed image compared with the other blocking effect reduction methods in existence.
