Spring 2012

Joint and Disjoint Demosaicing of RGB and NIR Images

The main goal of this project is to compare disjoint and joint demosaicing approaches for RGB and NIR images. To this end, first the algorithm proposed in [1] (the NEDI algorithm) is implemented to separately demosaic NIR and RGB images. In the next step, the color demosaicing algorithm of [2] (the frequency selection algorithm) is generalized to be used for RGB/NIR joint demosaicing. For fair comparison, instead of the bilinear interpolation used in [2], we will use the NEDI interpolation algorithm. At last the performance of NEDI and frequency selection algorithms will be compared. (If time allows, we will compute the correlation between color and NIR images. Using the correlation results and comparison results of NEDI and frequency selection algorithms, we might be able to come up with a systematic decision approach to switch between these algorithms to get higher reconstruction quality.)


[1] X. Li and M. T. Orchard, “New Edge-Directed Interpolation,” in IEEE Transactions on Image Processing, vol. 10, no. 10, 2001.
[2] D. Alleysson, S. Susstrunk, and J. Herault, “Linear Demosaicing Inspired by the Human Visual System,” in IEEE Transactions on Image Processing, vol. 14, no. 4, 2005.

Prerequisites: Good Matlab skills, basic knowledge about demosaicing & interpolation.

Type of Work: 50% theory, 50% programming.

Level: BS, semester project

Supervisor: Zahra Sadeghipoor (zahra.sadeghipoor@epfl.ch)