A novel efficient image compression system based on independent component analysis
Zafar Shahid, Florent Dupont, Atilla M. Baskurt · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2009
Next generation image compression system should be optimized the way human vision system (HVS) works. HVS has been evolved over millions of years for the images which exist in our environment. This idea is reinforced by the fact that sparse codes extracted from natural images resemble the primary visual cortex of HVS. We have introduced a novel technique in which basis functions trained by Independent Component Analysis (ICA) have been used to transform the image. ICA has been used to extract the independent features (basis functions) which are localized, bandlimited and oriented like HVS and resemble wavelet and Gabor bases. A greedy algorithm named matching pursuit (MP) has been used to transform the image in the ICA domain which is followed by quantization and multistage entropy coding. We have compared our codec with JPEG from the DCT family and JPEG2000 from the wavelets family. For fingerprint images, results are also compared with wavelet scalar quantization (WSQ) codec which has been especially tailored for this type of images. Our codec outperforms JPEG and WSQ and also performs comparable to JPEG2000 with lower complexity than the latter.