Machine learning for denoising
Quek, Kenneth Joo Hong · DR-NTU (Nanyang Technological University) · 2018
High image quality is desirable in fields like in the medical field where image analysis is often performed. With current technology, appearance of noises found on images are inevitable. There are many image denoising methods proposed with varying denoising results. Using a convolutional neural network that uses convolution layer, ReLU activation functions and Batch Normalization and training of 400 grayscale images we can denoise grayscale images of size180x180, 256x256 and 512x515 corrupted with additive white gaussian noise with different level of standard deviation applied. The experiment results show that the trained model can outperform different denoising methods like BM3D, EPLL and WNNM by 0.60 PSNR, 0.46 PSNR and 0.30 PSNR respectively.