In fractal image compression, most of the time during encoding is spent for finding the best matching pair of range-domain blocks. Different techniques have been analyzed for decreasing the number of operations required for this range-domain matching. Encoding time can be saved by reducing the domain search pool for each range block. Domain blocks can be classified based on local fractal dimension. Fractal dimension is being studied as a measure to analyze the complexity of image portions. This paper proposes application of height balanced binary search trees for storing domain information ordered in terms of the local fractal dimension. The approach is to prepare the domain pool dynamically, by comparing the fractal dimension of range block with that of the domains. Domains with fractal dimension in an interval, evenly covering the fractal dimension of range block alone are given for comparison. We use AVL trees to enlist the domains based on their fractal dimension. The domain pool is prepared at runtime. Since the tree organization is used in the preprocessing phase, the proposed method can be used with any algorithm for fractal compression.