Algorithme Hybride de Débruitage Image/Vidéo à base de Réseaux de Neurones Convolutionnels (CNN).

Abstract

Nowadays, Machine Learning has proven itself as a contender for modern researches on the technical domain. We propose in this manuscript to exploit a certain field from machine learning, which is the deep learning. With the Deep Learning we’ll do a study about image and Video denoising through deep neural network especially CNN (Convolutional Neural Network) according to several recent algorithms namely: DVDnet, FastDVDnet, FFDnet, DnCNN. We shall compare them with classical Image/Video denoising algorithms: BM3D, VBM4D, VNLB, SPTWO. By this comparative study, we will use an evaluation criterion by calculating PSNR, SSIM, MSE, and RMSE in addition to running time and visual quality of the obtained denoised images/videos.