Institutional Repository of Vilnius University
Neuroninių tinklų taikymas silpnai apšviestų vaizdų korekcijai /
Abstract
dc:descriptionRapid advances in technology make it possible to accurately capture the most important moments of everyday life, but capturing images accurately is subject to quality improvement processes. The quality of image capturing devices depends on environmental conditions such as low ambient light. As a result, captured images tend to lose the naturalness and accuracy of details. Traditional low-light image correction methods use histograms, masks or mathematical transformations which depend on statistical estimates and assumptions. However, each image requires a different amount of correction and tools. This work explores the application of neural networks in the field of low-light image correction and the solutions. The paper reviews the characteristics of noise artifacts generated during low-light image capture. A dataset of simulated illumination paired images in a composition of indoor interior and outdoor environments is constructed. The work develops a deep learning neural network model of the encoder-decoder architecture type, which is efficient in terms of execution time and correction accuracy, and can correct low-light images up to 20 frames per second. The work shows that training neural network models on a simulated illumination dataset is suitable for training the models but with a loss of accuracy for normally captured images.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Grublys, Evaldas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192846443&prefLang=en_US