MECHANISMS FOR IMPROVING THE QUALITY AND DENOISING OF IMAGES BASED ON THE CONVOLUTION AND RECURRENT NEURAL NETWORKS

Abstract

This article provides an overview of methods for image enhancement and denoising based on convolutional and recurrent neural networks with the addition of a non-local operations block. These methods are widely used in various domains. In medicine, these methods improve MRI images, assisting doctors in making accurate diagnoses. In security applications, these approaches enhance images and enable better visualization of details. The article covers the main existing approaches to image enhancement. The article presents an analysis of the key characteristics of the investigated neural networks, as well as the scenarios in which they are most effective. It also includes a table of results from several image enhancement methods and introduces a research method for comparing its effectiveness in image enhancement. The strengths of each approach are highlighted, and their efficiency in different scenarios is discussed. Considering specific characteristics of denoising tasks such as noise patterns, image types, and processing constraints can help in selecting the most suitable architecture to achieve the desired outcome. The article also highlights the use of the non-local operations block to improve image quality. This block is used to capture global dependencies among pixels, allowing better modeling of relationships between different parts of the image. The non-local operations block enables efficient detection of long-range dependencies and contextual information, leading to improved denoising and image restoration. Overall, this article is useful for researchers in the field of image processing and machine learning who are interested in understanding the key differences between convolutional neural networks (CNNs) and recurrent neural networks (RNNs) and exploring existing approaches to image enhancement and denoising. It provides a comprehensive overview of methods for image enhancement and denoising using convolutional and recurrent neural networks with the addition of a non-local operations block, along with information about existing approaches. The information and recommendations presented in this article can assist in selecting appropriate methods for addressing image processing tasks.