A new improved algorithm of image coding based on wavelet zero tree is proposed in this paper. Through the algorithm

the high frequency non-zero coefficients of wavelet transform are firstly predicted processed

then according to the principle of fractal coding

we predict other coefficients from the top-level of zero tree to the lower-level

and find out a series of the best matching range blocks that can represent all scales wavelet transform images. These range blocks are named representative blocks. They are composed representative trees. According to the representative trees

the decoder can reconstruct image by fractal predicting. Theoretical analysis and experimental results indicate that the improved algorithm of image coding based on wavelet zero tree in the paper can achieve a higher compression rate.