In this paper

a segmentation-based hybrid image coding algorithm is proposed which takes the features of the human vision system (HVS) into account. First

a novel criterion for quadtree-based segmentation using contrast masking is presented. This segmentation criterion performs better than the frequently used criterion

block variance. With this criterion

the original image is segmented into uniform blocks of 32×32

16×16 and blocks of 8×8 to be classified. Then a two-level classifier

based on contrast masking sensitivity and polynomial approximation error

is further applied to the blocks of 8×8 such that all blocks are classified into a few categories: uniform blocks of different sizes

smooth blocks of 8×8 and non-smooth blocks of 8×8. At last

block mean

polynomial approximation and DCT coding are applied to the blocks accordingly. The experimental results demonstrate that the proposed algorithm performs much better than JPEG.