Classified Block Coding has received more attention recently for the advantage of easy implementation. The classification and coding of blocks are the two important problems in classified block coding. In this paper

we propose a classified block coding algorithm based on DCT coding and polynomial approximation. In the algorithm

the original image is splitted into non-overlapped 8×8 blocks. The blocks are classified into four classes: constant blocks

smooth blocks

coarse blocks and detail blocks

by using the intensity local variance

the polynomial approximation error and spatial-frequency distribution. The constant blocks and smooth blocks areapproximated by 0-order and 1-order polynomial respectively. For the coarse and detail blocks

we compute and quantize their DCT coefficients. Then encode them by means of an improved run-length coding. The experiment results show that the proposed algorithm

without using entropy coder as postprocessor of the codes

has better performance than JPEG.