Abstract: This article gives an improved algorithm. The original image is directly divided into larger blocks D_i, and also divided into smaller blocks or the same size of R_i which are divided into smaller categories by the defined standard. And then it searches the attractor in the corresponding classes, if it does not find out the right R_i, it will construct the fictitious Ȓ_i and save the correctional value instead of searching R_i from the huge database. In this way, it will save most time and greatly improved the speed of the coding.