CNN based noise classification and denoising of images
Our goal is to find whether a convolutional neural network (CNN) performs better than the existing blind algorithms for image denoising, and, if yes, whether the noise statistics has an effect on the performance gap. We performed automatic identification of noise distribution, over a set of nine possible distributions, namely, Gaussian, log-normal, uniform, exponential, Poisson, salt and pepper, Rayleigh, speckle and Erlang. Next, for each of these noisy image sets, we compared the performance of FFDNet, a CNN based denoising method, with noise clinic, a blind denoising algorithm.
Denoising results for nine different types of noises. For each noise type, from left to right: original image, noisy image, blind denoising, CNN-based denoising. Images used: image 2092, 3096, 8023, 8049 and 12074 from BSDS300 dataset [gray].
For complete analysis refer to our conference paper: D. Sil, A. Dutta and A. Chandra. Convolutional neural networks for noise classification and denoising of images. In Proc. IEEE TENCON, pp. 447-451, Oct. 2019.