ASPECTS OF USING NEURAL NETWORKS TO IMPROVE THE QUALITY OF IMAGES

Abstract

The article presents an overview of image enhancement and noise reduction methods based on convolutional and recurrent neural networks with the addition of non-local operations blocks. These methods are widely used in various industries. In medicine, they help improve the quality of MRI images, which, in turn, contributes to the accuracy of doctors' diagnoses. In the field of security, these technologies enable image enhancement and detail enhancement. The article covers the main available approaches to image enhancement. The article analyzes the main characteristics of the considered neural networks, as well as the scenarios in which they demonstrate the greatest effectiveness. A table with the performance of different image enhancement methods is also presented, to which the investigated method is added to evaluate its effectiveness in image enhancement. The paper emphasizes the advantages of each of these approaches and their effectiveness in different conditions. Considering the specific features of the denoising task, such as the type of noise, image type, and processing constraints, will help select the most appropriate architecture to achieve the desired result. The article also discusses the use of the non-local operations block to improve image quality. This block serves to identify global interrelationships between pixels, which contributes to better modeling of relationships between different parts of the image. Thanks to the non-local operations block, it is possible to efficiently detect long-term dependencies and contextual information, which in turn leads to improved noise disaggregation and image restoration. The article offers a comprehensive overview of image enhancement and denoising methods that use convolutional and recurrent neural networks with the addition of a non-local operations block, and also provides information on existing approaches. The information and guidelines provided in this article can be helpful in choosing appropriate methods for solving image processing tasks. The article is useful for image processing and machine learning researchers who want to understand the main differences between convolutional neural networks (CNNs) and recurrent neural networks (RNN), as well as with already known approaches to improving and reducing noise in images.