The sparse representation of image is the key to the image coding technique in transform domain. In this paper

based on the analysis of the limitation of ridgelet function

a new construction method of multiscale ridgelet dictionary is firstly proposed. The speed of search of the most matching atoms is improved based on the tree structured organization of atoms. And then a tree structure multiscale ridgelet dictionary based matching pursuit image coding scheme is proposed. In the last the decomposed coefficients are adaptively quantized and encoded based on the distribution of coefficients and the rate-distortion analysis. Experiment results show that the new multiscale ridgelet dictionary can represent the images sparsely. On the other hand

the performances of the new method are shown to compare favorably against those of JPEG2000 scheme

especially at low bit rate.