Evolved transforms improve image compression

FEBRUARY 4, 2009--Many image-processing applications, including JPEG2000 and fingerprint compression, use wavelets. By redistributing an image's energy into a set of lower-resolution trend subimages, discrete wavelet transforms can significantly reduce the number of bits required for representation. Quantization (approximating a signal with a smaller number of bits) and thresholding (setting all but the largest transform values to zero) may achieve additional compression. However, these techniques introduce irreversible information loss, since the mean squared error (MSE) of reconstructed images increases proportionately with both techniques.

One approach that may circumvent this problem is evolutionary computation, which uses fitness-based selection and genetics-inspired operators to evolve solutions to a combinatorial optimization task. Researchers at the University of Alaska Anchorage and Wright-Patterson Air Force Base have developed a novel technique for optimizing compression that outperforms wavelets for data subject to quantization or thresholding. For more information, go to: http://spie.org/x33319.xml?highlight=x2410&ArticleID=x33319