WaveICA: A hybrid tool for image processing

The paper proposes a modified ICA that gives solutions close to wavelet subbands. The obtained ICs have GGD distribution and contain signal details, like wavelet subbands, but are statistically independent, like sources. The algorithm is derived from FastICA, by replacing the whitening stage by a Wavelet Transform with coefficient normalization and by initially loading the mixing matrix with the wavelet filter. Some examples of applications from image processing area are given: texture recognition, noncompressible bitstrings generating and image filtering.