In fractal image compression, block partition methods have critical effects on the compression ratios and the quality of images. Conventional block partition methods are not sufficient to take advantage of the characteristics of human visual system, because they considers only the spatial-domain. But because of the nature of the image signal and the mechanism of human vision, frequency-domain must also be considered. And it is difficult to parallelize the compression algorithms because they use adaptive block partition.
We employ wavelet transform to use the frequency domain informations of image. By applying wavelet transform as preprocessing step, we can divide an image into smaller subbands with different frequencies. For the intraframe coding, each subband is partitioned in a deterministic way, and then fractal still image coding is applied. For the interframe coding, blocks of two adjacent frames are compared. If some blocks were changed more than a given threshold, it is fractal coded. The blocks with little change are searched for their motion vectors.
To evaluate the performance of our algorithm, we compared two conventional fractal moving image compression systems with ours, using a standard test sequence of images. Performance metrics were compression ratio and peak signal to noise ratio. As a result, we achieved 20%∼25% increased compression ratio with the 0.5∼0.7 increased PSNR. We also tested other image sequences to show the performance of our system.
