The wavelet transform suffers from the limitation of poor directional selectivity. To overcome this disadvantage, a novel scheme with joint Contourlet and Bandelet transform is proposed. First the image is decomposed into different frequency subbands by the wavelet transform, and then the high frequency subband is further decomposed into directional subbands by directional filterbank. Next, the Bandelet transform is taken on each directional subband. The proposed multidirectional and multiscale transform with critical sampling can represent the geometrical features such as edges and texture more sparsely, which is of great benefit to image compression. Finally, the bit plane and contextual coding procedure of the JPEG2000 scheme are applied to quantized transform coefficients to realize image compression. Experiments show that the proposed algorithms evidently outperform JPEG2000 for the images with abundant edges and texture.
