Computational Color Constancy: From Pixel to Video with a Stop at Convolutional Neural Network
Qian, Yanlin (2020)
Qian, Yanlin
Tampere University
2020
Tieto- ja sähkötekniikan tohtoriohjelma - Doctoral Programme in Computing and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Väitöspäivä
2020-12-11
Julkaisun pysyvä osoite on
https://urn.fi/URN:ISBN:978-952-03-1791-1
https://urn.fi/URN:ISBN:978-952-03-1791-1
Tiivistelmä
This doctor thesis shows the author’s personal knowledge pieces about Color Constancy world.
We start from mathematical representation of image formation model, tell the substantial difference between “illumination” and “illuminant”, and make a brief introduction to prior arts from the classic GrayWorld to the latest deep neural networkbased approaches. We describe the performance metrics and the popular datasets for evaluating color constancy approaches in an unified setting. The main part of this thesis is about describing our contribution over three aspects of color constancy research – a static illumination estimation method called Grayness Index (GI), a deep net architecture C4 for single-frame color constancy, and two temporal color constancy methods RCC-Net and TCC-Net. We also release a largescale temporal color constancy dataset, for accommodating data driven temporal methods.
Finally we explore color constancy with the assistance of flash light and prove that flash light helps deliver finer spatial illumination map.
We start from mathematical representation of image formation model, tell the substantial difference between “illumination” and “illuminant”, and make a brief introduction to prior arts from the classic GrayWorld to the latest deep neural networkbased approaches. We describe the performance metrics and the popular datasets for evaluating color constancy approaches in an unified setting. The main part of this thesis is about describing our contribution over three aspects of color constancy research – a static illumination estimation method called Grayness Index (GI), a deep net architecture C4 for single-frame color constancy, and two temporal color constancy methods RCC-Net and TCC-Net. We also release a largescale temporal color constancy dataset, for accommodating data driven temporal methods.
Finally we explore color constancy with the assistance of flash light and prove that flash light helps deliver finer spatial illumination map.
Kokoelmat
- Väitöskirjat [4754]