Preliminary version: https://arxiv.org/abs/2003.00762
We propose a learning-based denoising method called FlashLight CNN (FLCNN) that implements a deep neural network for image denoising. The proposed approach is based on deep residual networks and inception networks and it is able to leverage many more parameters than residual networks alone for denoising grayscale images corrupted by additive white Gaussian noise (AWGN). FlashLight CNN demonstrates state of the art performance when compared quantitatively and visually with the current state of the art image denoising methods:
FlashLightCNN is made up two phases:warm up and boost phases,with a residual skip connection between the input and the output. The warmup phase uses only conventional convolutional layers and resembles a typical cnn. The boost phase on the other hand, uses much wider residual inception layers that rapidly increase the number of parameters of the network while avoiding the dimishing feature reuse that would come with it if only conventional convolutional layers would be employed.
The boost phase implements the customized residual inception layers that designed to maintain the learning capacity of great deep neural networks as show in the figure below
The models used for valuating can be downloaded here
[1] K. Dabov, A. Foi, V. Katkovnik, and K. Egiazar-ian, “Image denoising by sparse 3-d transform-domain collaborative filtering,”IEEE Transactions on ImageProcessing, vol. 16, no. 8, pp. 2080–2095, aug 2007.
[12] Kai Zhang, Wangmeng Zuo, and Lei Zhang, “FFD-Net: Toward a fast and flexible solution for CNN-based image denoising,”IEEE Transactions on ImageProcessing, vol. 27, no. 9, pp. 4608–4622, sep 2018.
[13] Kai Zhang, Wangmeng Zuo, Shuhang Gu, and LeiZhang, “Learning deep CNN denoiser prior for imagerestoration,” in2017 IEEE Conference on ComputerVision and Pattern Recognition (CVPR). jul 2017,IEEE.
[14] Wuzhen Shi, Feng Jiang, Shengping Zhang, Rui Wang,Debin Zhao, and Huiyu Zhou, “Hierarchical residuallearning for image denoising,”Signal Processing:Image Communication, vol. 76, pp. 243–251, aug 2019