A Fast C++ Implementation of Neural Network Backpropagation Training Algorithm: Application to Bayesian Optimal Image Demosaicing

Abstract

Recent years have seen a surge of interest in multilayer neural networks fueled by their successful applications in numerous image processing and computer vision tasks. In this article, we describe a C++ implementation of the stochastic gradient descent to train a multilayer neural network, where a fast and accurate acceleration of tanh(.) is achieved with linear interpolation. As an example of application, we present a neural network able to deliver state-of-the-art performance in image demosaicing. Source CodeThe source code and an online demo are accessible at the IPOL web page of this article 1 .

A Study of Two CNN Demosaicking Algorithms

“…A recently emerged trend consists in using convolutional neural networks for image processing and computer vision applications. See for example the early attempt in [19]. These methods have by now reached state-of-the-art results.…”

Small Neural Networks can Denoise Image Textures Well: a Useful Complement to BM3D

“…We do not detail the training of our denoising neural networks as it has already been documented [7]. The only difference is that following [2], we collected the training and validation patches from Pascal VOC2012 2 and processed them according to Algorithm 1 before feeding them to the neural networks.…”

“…For example, it has been shown that a multilayer neural network, when endowed with sufficient capacity, could be trained to deliver stateof-the-art performance in denoising [2,8] as well as in demosaicing [5,9,7].…”