Abstract: An image transformation method to reduce the number of the gray levels is disclosed. Error diffusion, error bias and a programmable integer C are utilized. For the focus pixel, the value Gi of the original gray scale is read. The related errors Ej previously obtained from some chosen pixels are loaded in. The respective corresponding weighting coefficient Wj for each errors Ej are determined. Then, a resultant gray value Ri of the focus pixel is calculated by Ri=Gi+&Sgr;Wj(Ej−C), wherein C is a programmable integer. Thereafter, the gray value Ri is dividing into most significant part Mi and least significant part Li, wherein Mi is a K-bit integer, Li is an (N−K)-bit integer, Mi×2N−K+Li=Ri, and −2N−K/2≦Li<2N−K/2. The error Ei of the focus pixel is stored as Li plus an error bias D of 2N−K/2. Finally, the focus pixel is displayed using Mi as the degraded target gray value.