Abstract: There is a Poisson inverse problem in biomedical imaging, fluorescence microscopy and so on. Sincethe observed measurements are damaged by a linear operator and further destroyed by Poisson noise, recoveringthe approximate original image is difficult. Motivated by the decouple scheme and the variance-stabilizing transformation(VST) strategy, we propose a method of transformed convolutional neural network (CNN) to restorethe observed image. In the network, the Conv-layers play the role of a linear inverse filter and the distributiontransformation simultaneously. Furthermore, there is no batch normalization (BN) layer in the residual block ofthe network, which is devoted to tackling with the non-Gaussian recovery procedure. The proposed method iscompared with state-of-the-art Poisson deblurring algorithms, and the experimental results show the effectivenessof the method.

Key words: deconvolution, Poisson noise, transformed network, decouple scheme, variance-stabilizing transformation(VST)