Phase-only hologram denoising based on attention wavelet residual neural network

Abstract

In order to suppress the interference of speckle noise on the quality of hologram an attention-wavelet residual denoising network based on multi-level wavelet convolutional neural network is proposed. Firstly, the phase-only hologram is input into the network, and the wavelet transform of the shrinkage subnet is used instead of pooling to reduce the size of the hologram and expand the receptive field. Secondly, the attention-residual module which embeds the convolutional block attention module(CBAM) into the residual block is added to each layer of the network. The CBAM is used to capture the speckle noise of the hologram more accurately, and to improve the fusion capability of network feature information through the residual block. Finally, the hologram after denoising is obtained by up-sampling in the extended subnet by inverse wavelet transform. Simulation experiments show that the improved network effectively removes speckle noise while obtaining high quality holograms.