On existing methods for removing noise from an image

Abstract

On existing methods for removing noise from an image

Moseva M.S., Kharrasov K.R.

Incoming article date: 24.06.2023

This paper considers existing classical and neural network methods for combating noise in computer vision systems. Despite the fact that neural network classifiers demonstrate high accuracy, it is not possible to achieve stability on noisy data. Methods for improving an image based on a bilateral filter, a histogram of oriented gradients, integration of filters with Retinex, a gamma-normal model, a combination of a dark channel with various tools, as well as changes in the architecture of convolutional neural networks by modifying or replacing its components and the applicability of ensembles of neural networks are considered.

Keywords: image processing, image filtering, machine vision, pattern recognition