EE Seminar: Example Based Demosaicing

~~Speaker: Gilad Michael, 
M.Sc. student under the supervision of Prof. Nahum Kiryati

Monday, July 6, 2015 at 15:00
Room 011, Kitot Bldg., Faculty of Engineering

Example Based Demosaicing

Abstract

Demosaicing is an algorithm used to reconstruct a color image from the incomplete color samples of a color filter array (CFA). Simple interpolation approaches for demosaicing tend to produce artifacts or loss of resolution mainly due to mistreating chroma information as luma or vice versa. Despite significant progress in the past decade, current state of the art demosaicing algorithms still tend to produce artifacts at high-saturation edges. In this paper we propose a new approach to demosaicing - example based.
We suggest using a dictionary which holds pairs of a Bayer image patch and the corresponding full resolution RGB patch. Demosaicing is performed by finding a matching dictionary entry with similar Bayer image patch and use entry's RGB counterpart is to estimate the full resolution color image. Unlike previous Comparative experimental evaluation shows that example-based demosaicing (EBD) produces visually superior, artifact-free results.