On Luminance Noise Removal Using Convolutional Neural Network

This paper's focus is the problem of removing a luminance noise from a raster image, more precisely - an application of artificial neural networks to luminance noise removal.A neural network is one of the newest approaches to image processing, that began to gain popularity nearly in 2010. Nowadays noise removal via neural network processing tends to replace more traditional methods, such as linear and nonlinear filtering. This paper compares two algorithms of removing luminance noise from grayscale images: passing an image into the neural network to remove noise directly, and applying a Gaussian blur filter to an image and passing the blurred one into the neural network to restore the sharpness. The result of a comparative analysis is, that a method with applying Gaussian blur to the image and subsequent image restoring shows better image quality and PSNR metric results, than direct noise removal via a neural network.