Naslov
Real-time Retinex-based and learning-based methods for computational color constancy with high accuracy

Autori
Banić, Nikola

Vrsta, podvrsta i kategorija rada
Ocjenski radovi, doktorska disertacija

Fakultet
Fakultet elektrotehnike i računarstva

Mjesto
Zagreb

Datum
15.09

Godina
2016

Stranica
104

Mentor
Lončarić, Sven

Ključne riječi
image processing; color constancy; white balance

Sažetak
The doctoral thesis of Nikola Banić, mag. ing. comp., entitled Real-time Retinex-based and learning-based methods for computational color constancy with high accuracy, describes his research results in the field of image processing in the area of computational color constancy (CCC). CCC methods are used to remove the influence of illumination and camera characteristics on image colors. This is important for displaying images, but it is especially important for further computational analysis of images, where color information usually plays an important role. Such methods have to be implemented in all modern cameras as a part of their built in image processing pipeline. For this reason, low computational complexity is an important characteristic, besides high accuracy. Such a compromise is difficult to achieve which makes CCC an important research topic. CCC methods are based on various color constancy models but so far none of them managed fully to explain how the human visual system achieves color constancy. Human visual system depends on both global and local color allocation, hence some CCC methods are designed as local methods, and some are designed as global methods. Most local methods can be extended to perform global CCC. Also global CCC methods are usually part of cameras built in image processing pipeline, while local CCC algorithms are usually applied later on as part of an image processing software. The presented research focuses on local and global CCC methods for estimation of illumination and tries to achieve optimal compromise between accuracy and low computational complexity. Global illumination estimation methods can be divided in two groups: low-level statistics-based ones and learning-based ones. The low-level statistics-based methods are usually faster, which is the reason why most digital cameras use them. On the other hand, the learning- based methods are usually more accurate. The author presents research results in both groups of global CCC.

Izvorni jezik
Engleski

Znanstvena područja
Elektrotehnika, Računarstvo

POVEZANOST RADA