Fractal image compression using component analysis networks.

摘要

The partitioned iterated function systems (PIFS) fractal image compression provides very competitive rate-distortion curves and fast decoding. However, it suffers from long encoding time. So far, several methods have been proposed in order to reduce the time complexity of the encoding.; In this thesis, three novel neural network techniques, mixture of non-linear principal components (MNLPC), mixture of independent components (MIC) and high-dimensional mixture principal components (H-MPC) are developed to reduce the encoding complexity of the PIFS fractal coding.; Applying these novel techniques, the potential best range-domain block matches in the PIFS encoding phase are confined to some relatively small size domain block pools, i.e. network libraries. The encoding time is shortened dramatically using the proposed techniques. The experimental results also demonstrate the new methods' compression performances are better than that of the standard PIFS coding and that of the PIFS coding using the MPC network library. H-MPC networks can provide minimal distortion, while MNLPC and MIC networks can achieved high compression ratio.