Mid Sweden University

miun.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Comparative Study of Deep Learning-based Generative Models for Image Compression
Mid Sweden University, Faculty of Science, Technology and Media, Department of Computer and Electrical Engineering (2023-).
2024 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
Abstract [sv]

Denna studie fördjupar sig i effektiviteten av generativa modeller, nämligen Variational Autoencoders (VAEs), Diffusion Models (DMs) och Generative Adversarial Networks (GANs), i bildkomprimering. Forskningen fokuserar på att utvärdera dessa modeller utifrån deras förmåga att komprimera bilder med bibehållen visuell trohet. Utvärderingsmått som Peak Signal-to-Noise Ratio (PSNR) och Bits Per Pixel (bpp) används för att bedöma modellernas prestanda. Genom en jämförelse av olika tillvägagångssätt och förtränade modeller identifierades en praktisk och effektiv metod för bildkomprimering som belyste potentialen hos generativa modeller inom detta område. Metoden som användes var att låta modellerna komprimera en uppsättning av fem bilder från kodak-dataset, de komprimerade bilderna var desamma längs alla modeller. De rekonstruerade bilderna analyserades sedan genom att mäta deras Peak signal-to-noise ratio (PSNR) och värdet för bitar per pixel (BPP) för varje modell. Resultaten visar att VAE tenderar att ge bättre återgivningsbilder med ett PSNR-medelvärde 37 jämfört med GAN(28) och DM(30) medan GAN:er minskar bildstorleken mest med och ett medelvärde på 0,2 bitar per pixel jämfört med 0,7 för DM och 0,9 för VAE.

Abstract [en]

This study delves into the effectiveness of generative models, namely Variational Autoencoders (VAEs), Diffusion Models (DMs) and Generative Adversarial Networks (GANs), in image compression. The research focuses on evaluating these models based on their ability to compress images while maintaining visual fidelity. Evaluation metrics such as the Peak Signal-to-Noise Ratio (PSNR) and Bits Per Pixel (bpp) are utilized to assess the performance of the models. Through a comparison of different approaches and pre-trained models, a practical and efficient method for image compression was identified, shedding light on the potential of generative models in this domain. The method used was letting the models compress a set of five images from the kodak dataset, the images compressed were the same along all models. The reconstructed images were then analyzed by measuring their their Peak signalto-noise ratio(PSNR) and the bits per pixel(BPP) value of each model. The results show that the VAE tends to give better fidelity images with a PSNR average value 37 compared to GAN(28) and DM(30) while GANs reduce the image size the most with and average value of 0.2 bits per pixel compared with 0.7 for DM and 0.9 for VAE.

Place, publisher, year, edition, pages
2024. , p. 41
Keywords [en]
BPP, PSNR, GAN, VAE, DM
Keywords [sv]
BPP, PSNR, GAN, VAE, DM
National Category
Software Engineering
Identifiers
URN: urn:nbn:se:miun:diva-51615Local ID: DT-V24-G3-007OAI: oai:DiVA.org:miun-51615DiVA, id: diva2:1874497
Subject / course
Computer Engineering DT1
Educational program
Master of Science in Engineering - Computer Engineering TDTEA 300 higher education credits
Supervisors
Examiners
Available from: 2024-06-20 Created: 2024-06-20 Last updated: 2024-06-20Bibliographically approved

Open Access in DiVA

fulltext(5424 kB)238 downloads
File information
File name FULLTEXT01.pdfFile size 5424 kBChecksum SHA-512
100a7b6490d310faf7fdb1cf2a1768d0f394025ca8d71847d28d7cb889a8f4f732a81f0aa7e56947e1acb661b140f190ae96834ad3bb939efb28cd752bc2db0c
Type fulltextMimetype application/pdf

Search in DiVA

By author/editor
Kanda, Isaac Ntambu
By organisation
Department of Computer and Electrical Engineering (2023-)
Software Engineering

Search outside of DiVA

GoogleGoogle Scholar
Total: 238 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 216 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf