An evolutionary self-organizing neural network for blind deconvolution

We propose an evolutionary self-organizing neural network for blind deconvolution. The evolutionary self-organizing neural network has two steps of learning: one is the self-organizing learning, the other focuses on a genetic algorithm. The self-organizing learning function maximizes the Gibbs distribution of the information distance. An improvement by the genetic algorithm is made between two iterations of self-organizing learning. The learning gradually reduces the information distance of a model and a degraded image into a global minimum. We compare the blind deconvolution results by the proposed neural network with those by the Richardson-Lucy algorithm that is widely used in blind deconvolution. The computer simulations demonstrate that the evolutionary self-organizing learning converges faster, and gives good reconstruction as well. Also the evolutionary self-organizing blind deconvolution algorithm is found to be more effective and insensitive to image noise