A new adaptive transform image coding

which is based on the analysis of features for image edges and brain mechanism of vision

is presented in this paper. The image sub-blocks are classified according to the control information of the image edges in spatial domain firstly

then resampling

quantizing and coding are realized according to different features of these classifications in transform domain

particularly some rectangle filters are de-signed suitable to data compression of the image sub-blocks which are part of the lines and directional edges in the image. Finally

the results of computer simulation with the example of a standard image of human head and shoulder are given. The experimental results show that the subjective quality of the coding image is good and the PSNR reaches 33.9dB when the code rate is as low as 0.2bit/pel.