Deep convolutional neural network for image deconvolution
L.Xu, J.S.Ren, C.Liu and J.Jia, Advances in Neural Information, NIPS, 2014
MATLAB
Deep convolutional neural network comprises of two modules to remove blurs,artifacts and noise. The first module DCNN(Deconvolution Convolutional Neural Network) is trained with separable kernel inversion with 2 hidden layers.The output of this module serves as the input to the second module, it is otherwise called as denoise CNN module to remove outliers. It also 2 hidden layers with sigmoid or tanh as the activation function. Both the modules are trained in supervised fashion with proper initialization.
Deep convolutional neural network comprises of two modules to remove blurs,artifacts and noise. The first module DCNN(Deconvolution Convolutional Neural Network) is trained with separable kernel inversion with 2 hidden layers.The output of this module serves as the input to the second module, it is otherwise called as denoise CNN module to remove outliers. It also 2 hidden layers with sigmoid or tanh as the activation function. Both the modules are trained in supervised fashion with proper initialization.