Izboljšanje kvalitete videoposnetkov v realnem času

magistrsko delo

Šimen Ravnik (Author), Matija Marolt (Mentor)

Abstract

V magistrskem delu predstavimo globoki model za super-ločljivost videoposnetkov, ki omogoča izboljšanje ločljivosti slike v realnem času. Predlagana arhitektura vključuje tri glavne komponente: 2D konvolucijski modul in transformerski modul za izluščanje prostorskih značilk ter prilagojeno arhitekturo modela BasicVSR za izluščanje časovnih odvisnosti med okvirji videoposnetka. Ključni prispevek dela je vpeljava transformerskega modula v arhitekturo modelov za super-ločljivost videoposnetkov. Uporabili smo tehniko razvijanja za pretvorbo vhodne slike v 1D sekvenco, ki služi kot vhod v transformer. To nam omogoča zajem dolgoročnih odvisnosti znotraj slike, ki so lahko ključne za samo rekonstrukcijo. Rezultati so pokazali, da naš model dosega zadovoljive rezultate v primerjavi s trenutno uveljavljenimi modeli za super-ločljivost videoposnetkov, pri čemer je bil dosežen boljši čas izvajanja. Kljub višji zahtevi po pomnilniku je naš model uspešno izboljšal vizualno kakovost slik v realnem času. Poudarili smo tudi, da visoke vrednosti PSNR in SSIM niso vedno najboljši pokazatelji kakovosti slike, saj je pri oceni rezultatov pomembna tudi vizualna ocena.

Keywords

super ločljivost videoposnetkov;odstranjevanje šuma;sistem v realnem času;kvaliteta videoposnetkov;računalništvo in informatika;magisteriji;

Data

Language:	Slovenian
Year of publishing:	2023
Typology:	2.09 - Master's Thesis
Organization:	UL FRI - Faculty of Computer and Information Science
Publisher:	[Š. Ravnik]
UDC:	004:621.397(043.2)
COBISS:	181728003
Views:	60
Downloads:	14
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Real-time video super-resolution
Secondary abstract:	In this work, we present a deep learning model for video super-resolution that allows real-time video quality enhancement. The proposed architecture includes three main components: 2D convolutional module and transformer module for spatial feature extraction, and customized architecture of the BasicVSR model for extracting temporal dependencies between video frames. The key contribution of this work is the introduction of the transformer module into the architecture of video super-resolution models. We used unfolding technique to convert the input image into a 1D sequence, which serves as input to the transformer. This enables us to capture long-term dependencies within the image, which can be crucial for the reconstruction itself. The results have shown that our model achieves satisfactory results compared to currently established models for video super-resolution, with improved execution time. Despite the higher memory requirement, our model successfully enhances the visual quality of videos in real-time. We also emphasized that high PSNR and SSIM values are not always the best indicators of image quality, as visual evaluation is also important for assessing the results.
Secondary keywords:	videos;video super resolution;video denoising;real-time system;computer science;computer and information science;master's degree;Videoposnetki;Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS):	Master's thesis/paper
Study programme:	1000471
Embargo end date (OpenAIRE):	1970-01-01
Thesis comment:	Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages:	53 str.
ID:	22322770