Publication result detail

On the Visual Quality of AI and non-AI Images Compressed by Different Autoencoders

SAMOFALOV, A.; POLÁK, L.

Original Title

On the Visual Quality of AI and non-AI Images Compressed by Different Autoencoders

English Title

On the Visual Quality of AI and non-AI Images Compressed by Different Autoencoders

Type

Paper in proceedings (conference paper)

Original Abstract

Image compression using deep learning (DL) tech- niques is an emerging and rapidly evolving field. This approach has the potential to enhance image compression by learning complex patterns and representations from data, enabling higher compression ratios while maintaining high image quality. Unlike conventional compression methods, which rely on predefined algorithms, DL models can adapt and optimize compression based on the specific content of an image. This paper pro vides a comparison-based study of the two autoencoder models (dense and convolutional), commonly used in DL models for image compression. The comparison is based on objective metrics applied to human-made images from a publicly available database and AI-generated images to evaluate the quality of compressed images. The results obtained show that the autoencoders differ in terms of the visual quality of the reconstructed images.

English abstract

Image compression using deep learning (DL) tech- niques is an emerging and rapidly evolving field. This approach has the potential to enhance image compression by learning complex patterns and representations from data, enabling higher compression ratios while maintaining high image quality. Unlike conventional compression methods, which rely on predefined algorithms, DL models can adapt and optimize compression based on the specific content of an image. This paper pro vides a comparison-based study of the two autoencoder models (dense and convolutional), commonly used in DL models for image compression. The comparison is based on objective metrics applied to human-made images from a publicly available database and AI-generated images to evaluate the quality of compressed images. The results obtained show that the autoencoders differ in terms of the visual quality of the reconstructed images.

Keywords

image compression, deep learning, autoencoder, AI-generated image, objective metric, image quality

Key words in English

image compression, deep learning, autoencoder, AI-generated image, objective metric, image quality

Authors

SAMOFALOV, A.; POLÁK, L.

Released

12.05.2025

ISBN

979-8-3315-4447-8

Book

35th International Conference Radioelektronika (RADIOELEKTRONIKA)

Pages from

1

Pages to

5

Pages count

5

URL

https://ieeexplore.ieee.org/abstract/document/11008397

BibTex

@inproceedings{BUT197862,
  author="Andrii {Samofalov} and Ladislav {Polák}",
  title="On the Visual Quality of AI and non-AI Images Compressed by Different Autoencoders",
  booktitle="35th International Conference Radioelektronika (RADIOELEKTRONIKA)",
  year="2025",
  pages="1--5",
  doi="10.1109/RADIOELEKTRONIKA65656.2025.11008397",
  isbn="979-8-3315-4447-8",
  url="https://ieeexplore.ieee.org/abstract/document/11008397"
}