Evolved transforms for image reconstruction

This investigation uses a genetic algorithm to optimize coefficient sets describing inverse transforms that significantly reduce mean squared error of reconstructed images. Quantization error introduced during image compression and reconstruction is one of the worst noise sources, due to the fact that information is always permanently lost during the transformation process. Our approach establishes an adaptive filtering methodology for evolving transforms that outperform discrete wavelet inverse transforms for the reconstruction of images subjected to quantization error. Inverse transforms evolved against a single training image consistently generalize to exhibit superior performance against other images from the test set.