Reduction of Domain Blocks Based on a Clustering Method for Fractal Compression

Abstract

The process of fractal compression generates a number of intermediate artifacts during the encoding phase. These artifacts, known as range blocks and domain blocks, largely determine the coding efficiency. The coding efficiency can be characterized by the processing time required and the quality of the reconstructed image. A reduction in the number of domain blocks is very beneficial to the coding efficiency. The number of domain blocks is a crucial factor when reconstructing an image of high quality. However, deletion of too many domain blocks or simply some blocks with important characteristics can have very negative effect on the quality of reconstructed images. To obtain a good reconstructed image, it must be ensured that the deleted domain block do not carry any important characteristics and should be identified as insignificant before deletion. We apply the clustering method here to group related blocks together to ensure that the deleted blocks are redundant. That is, the retained blocks are significantly different to each other. Thus, the range blocks are compared only to significant domain blocks and not to redundant ones. The reduction of the intermediate data greatly benefits coding efficiency as well. The experimental results show that the peak signal-to-noise ratio (PSNR) of the reconstructed image is close to or around 30dB, that is, the quality of the reconstructed image is visually as good as the original.