A new fractal predictive image coding scheme based on zerotrees of wavelet coefficients is presented. We first pick out the range blocks to be predicted based on the zerotrees from subimages of the same orientation but different resolutions. These range blocks will form a range tree

something like zerotrees. For each range block

we then look for its optimum fractally-matched domain block in the subimage of the same orientation but lower resolution than that of the range block. Each domain block obtained this way can extend through different resolutions along the same orientation to form a domain tree. Among all these domain trees

one closest to the range tree will be its predictive tree. Theoretical analysis and experimental results presented in the paper show that our scheme can obtain higher compression rate at the little cost of signal-to-noise ratio.