Blind Separation for Blurred Images Based on the Adaptive Nonholonomic Natural Gradient Algorithm

A blind separation algorithm for restoring the original images from the blurred grayscale images is proposed, which utilizes the constrain ability of the nonholonomic natural gradient (NNG) in the independent component analysis(ICA) methods. However, the nonlinear activation function of this algorithm relates to the unavailable probability distribution of the sources closely, though it is robust to nonstationary and strongly undulate sources. To this problem, our method adaptively select the nonlinear function by use of the kurtosis of the output signals, and propose an adaptive NNG (ANNG) blind separation algorithm of blurred image based on ICA, and research the effect of the different mixture matrices to the performance of this algorithm. The simulations show the validity of the proposed method. Compared with the nonholonomic natural gradient algorithm and the classical FastICA algorithm, the performance index of this paper algorithm is also better.