Naslov
Noise removal from images by applying deep neural networks

Autori
Ivezić, Dijana ; Livada, Časlav

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
31st International Conference on Organization and Technology of Maintenance (OTO 2022) / - , 2022, 1-17

Skup
31st International Conference on Organization and Technology of Maintenance (OTO 2022)

Mjesto i datum
Osijek, Hrvatska, 12.12.2022. - 13.12.2022

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
neural networks ; noise removal ; Keras

Sažetak
This article addresses the problem of removing noise from images using neural networks. The type of noise that is removed in this article is salt and pepper noise. This noise represents jumps in intensity in the image, usually caused by data transmission errors. To quantify the success of the noise removal model, PSNR metrics are used. The paper provides a theoretical basis for the task in the form of a description of the neural network and its components, as well as its operation. It also describes deep learning and the differences and advantages of each type. The neural network used in this paper is based on the U-Net architecture. Using the Python programming language and the Keras application user interface, a model was developed to remove added white Gaussian noise and salt and pepper noise of varying intensity from images in the CIFAR-10 and MNIST image databases. The model is applied to the author's image that is not included in the tested databases. The analysis is presented in the PSNR presented for different noise intensities and examples of images with and without noise.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA