In recent years, several methods to solve the colour constancy problem have been introduced and studied. Colour constancy is an important area in machine vision: it provides a visual system with the capability to compensate for the effect of illumination in a scene. The colours we humans perceive are not easily expressible by a physical apparatus, which is in fact more sensitive to changes in illumination conditions. The human visual system is able to make several compensations and adjustments to ambient illumination conditions, so that we perceive illuminant-independent descriptors of the scene.
This thesis represents a series of experimental approaches to the colour constancy problem. An important part focuses on discussions and concrete implementations of the well-known and controversial Retinex model for human vision and application of the method on real images and choosing its parameters. The Retinex theory of human vision represents an important contribution in colour vision with strong implications in the colour constancy problem. Although the theory has been around for more than three decades, the lack of an efficient implementation and analysis of the effect of its parameters has raised many discussions and scepticism. This part of the thesis provides some common ground in further investigations into the Retinex theory.
A novel method for acquiring a large database for colour constancy research is presented, along with direct applications with improvements to the colour by correlation method. A large image database for colour constancy provides sufficient data to compute some meaningful statistics with respect to the interaction between the colours observed in the world and the actual measured illuminants in which these colours are recorded. The use of these statistics in the Bayesian approach improves the performance of the colour by correlation method for colour constancy. Another application is in the form of testing the effectiveness on real images of a recently introduced theory for colour constancy using the redness-luminance correlation in scenes.
[1]
Michael J. Swain,et al.
Color indexing
,
1991,
International Journal of Computer Vision.
[2]
James W. Jenness.
Human Color Vision (Second Edition)
,
1997
.
[3]
Graham D. Finlayson,et al.
Color in Perspective
,
1996,
IEEE Trans. Pattern Anal. Mach. Intell..
[4]
A Hurlbert,et al.
Formal connections between lightness algorithms.
,
1986,
Journal of the Optical Society of America. A, Optics and image science.
[5]
Jean Ponce,et al.
Computer Vision: A Modern Approach
,
2002
.
[6]
D H Hubel,et al.
Brain mechanisms of vision.
,
1979,
Scientific American.
[7]
B. Wandell.
Foundations of vision
,
1995
.
[8]
Alessandro Rizzi,et al.
Color appearance approach to image database visual retrieval
,
1999,
Electronic Imaging.
[9]
Laurence T Maloney,et al.
Illuminant estimation as cue combination.
,
2002,
Journal of vision.
[10]
John J. McCann,et al.
Lessons Learned from Mondrians Applied to Real Images and Color Gamuts
,
1999,
CIC.
[11]
Michael H. Brill,et al.
Color appearance models
,
1998
.