VGG-19 guided colour distribution matching
Ciobanu, Cristian (2025)
Kandidaatintyö
2025
School of Engineering Science, Tietotekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe20251209116596
https://urn.fi/URN:NBN:fi-fe20251209116596
Tiivistelmä
Traditional colour grading approaches struggle with consistency across varying lighting conditions, colour profiles, and content types.
The thesis investigates whether a CNN can be implemented for a more constant colour grading process. The integration of a modified pre-trained model in a transfer learning environment is explored. A design science research approach is taken for the creation of an artefact.
The findings include both quantitative and qualitative results. Performance is evaluated using Delta-E, SSIM, and PSNR benchmarks to compare colour and content accuracy. The results showcase the trade-offs and advantages of CNN-powered solutions in comparison with traditional approaches. The artefact outperforms the traditional methods in 2 out of 3 benchmarks.
The thesis concludes that a CNN-powered solution can perform a photorealistic colour transfer. Limitations, future work, and recommendations on the proper use of the artefact are provided. The target uses and other potential application areas are also highlighted.
The thesis investigates whether a CNN can be implemented for a more constant colour grading process. The integration of a modified pre-trained model in a transfer learning environment is explored. A design science research approach is taken for the creation of an artefact.
The findings include both quantitative and qualitative results. Performance is evaluated using Delta-E, SSIM, and PSNR benchmarks to compare colour and content accuracy. The results showcase the trade-offs and advantages of CNN-powered solutions in comparison with traditional approaches. The artefact outperforms the traditional methods in 2 out of 3 benchmarks.
The thesis concludes that a CNN-powered solution can perform a photorealistic colour transfer. Limitations, future work, and recommendations on the proper use of the artefact are provided. The target uses and other potential application areas are also highlighted.