Revolutionary image compression and reconstruction via evolutionary computation, part 2: multiresolution analysis transforms
Frank W. Moore, Brendan J. Babb · 2006
Abstract:- Previous research demonstrated that a genetic algorithm (GA) can utilize supercomputers to evolve image compression and reconstruction transforms that reduce mean squared error (MSE) by more than 22% (1.126 dB) under conditions subject to quantization, while continuing to average the same amount of compression as the Daubechies-4 (D4) wavelet. This paper describes subsequent research that extends our GA to evolve multi-resolution analysis (MRA) transforms. Test results indicate that our evolved MRA transforms can reduce MSE by an average of more than 10 % (0.50 dB) at three levels of decomposition. This result substantially improves upon state-of-the-art MRA transforms for compression and reconstruction applications subject to quantization error. Key-Words:- wavelets, genetic algorithms, image compression, quantization, multiresolution analysis