Kompresija video posnetkov z nevronskimi mrežami

Pelicon, Jan (author), Čehovin Zajc, Luka (mentor)

Diplomsko delo obravnava video kompresijo z uporabo nevronskih mrež. V zadnjih letih se je namreč z napredkom strojnega učenja pojavila ideja, da bi se kompresijo slikovnih in video podatkov lahko naučili z ustrezno arhitekturo nevronske mreže in veliko količino učnih podatkov. V nalogi smo se osredotočili na uporabo konvolucijskih samokodirnikov, ki slikovne podatke iz vhodnega prostora preslikajo v bolj kompakten latentni prostor ter nazaj. Predstavimo dva pristopa za kompresijo podatkov, prvi ima za cilj zgolj kompresijo posameznih slik, drugi pa predstavlja nadgradnjo v smeri video kompresije, ki sledi klasičnemu pristopu napovedovanja gibanja delov slike ter kodiranju popravkov. Opisali smo uporabljene arhitekture ter postopek učenja in testiranja. Več pozornosti smo posvetili operaciji kvantizacije, ki je pomemben element preko katerega kontroliramo nivo kompresije in kvaliteto rekonstrukcije. Testirali smo osnovno implementacijo in primerjali zmogljivost v primerjavi z JPEG formatom. Za testiranje druge implementacije smo si izbrali dve konfiguraciji, ju testirali pri različnih parametrih in primerjali s standardnimi kodeki za video kompresijo. Čeprav sta oba pristopa učinkovito kompresirala podatke, nista dosegala trenutnih standardov, zato predstavimo možne izboljšave, s katerimi bi se približali trenutnim standardom.

• Language: Slovenian
• Keywords: video kompresija, konvolucijske nevronske mreže, samokodirniki, kompresijsko ogrodje
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2020
• PID: 20.500.12556/RUL-119837
• COBISS.SI-ID: 30808835
• Publication date in RUL: 11.09.2020

PELICON, Jan, 2020, Kompresija video posnetkov z nevronskimi mrežami [online]. Bachelor’s thesis. [Accessed 24 March 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=119837

Secondary language

• Language: English
• Title: Video compression using neural networks

Abstract

This bachelor's thesis uses neural networks to compress video. Due to improvements in deep learning, a new idea appeared. Neural networks can learn to compress image and video data using large training sets and appropriate architecture. In the thesis, we used convolutional autoencoders that can transform input data into smaller latent space. We present two approaches to compression. The first one is designed to compress images, while the second is improved to compress video material. It is based on the classic approach of predicting movement in a scene and has error correction. We described used architectures and processes of learning and testing. We focused more on a quantization operation which is an important element for controlling compression ratio and quality. We evaluated the first approach and compared it with the JPEG image compression format. We chose two different configurations for the second approach, tested them using multiple parameters, and compared results with performances of standard codecs. Although both approaches are capable of efficient compression, they can not compete with today's standards. Because of this, we also mentioned some novelties that could significantly improve performance.

• Keywords: video compression, convolutional neural networks, autoencoders, compression framework