Color-filter-array (CFA) Demosaicking

Eric Dubois

This web site summarizes and provides links to my work on demosaicking for color filter arrays.

• E. Dubois, "Frequency domain methods for demosaicking of Bayer-sampled color images," IEEE Signal Processing Letters, vol. 12, pp. 847-850, Dec. 2005. (IEEE Xplore citation) This paper presented the first version of the adaptive luma-chroma demultiplexing algorithm for the Bayer CFA. The method gave competitive performance to the best methods available at the time. The algorithm used 21 by 21 FIR filters designed with the window method; thus it had relatively high computational complexity. There is a web page for the paper with more complete results.

• E. Dubois, "Filter design for adaptive frequency-domain Bayer demosaicking," Proc. IEEE Int. Conf. Image Processing (Atlanta GA), pp. 2705-2708, Oct. 2006. (IEEE Xplore citation) This paper introduces a least-squares methodology to design the filters for the luma-chroma demultiplexing algorithm, and was called LSLCD in later papers. It provided better performance than the system with the 21 by 21 window-designed filters, with much lower complexity. The maximum filter size beyond which no further improvement was obtained was 11 by 11. This method gave a state-of-the-art performance-complexity tradeoff. View the poster.

• M. Beermann and E. Dubois, "Improved demosaicking in the frequency domain by restoration filtering of the LCC bands," Visual Communications and Image Processing (San Jose CA), pp. 68221O: 1-10, Jan. 2008. (SPIE Digital Library) In his postdoctoral work, Marcus Beermann incorporated restoration filtering for the luma and chroma bands using bilateral filters, leading to improved demosaicking performance.

• E. Dubois, "Color filter array sampling of color images: Frequency-domain analysis and associated demosaicking algorithms," in Single-Sensor Imaging: Methods and Applications for Digital Cameras (R. Lukac, ed.), ch. 7, pp. 183-212, CRC Press, 2009. CRC Press catalog, CRCnetBASE. This book chapter presents a general theory for analysis and design with arbitrary periodic color filter array patterns, including non-rectangular structures. There is an associated website with additional details, extensive results, and software to reproduce the results for a number of the examples in the chapter.

• B. Leung, G. Jeon, and E. Dubois, "Least-squares luma-chroma demultiplexing algorithm for Bayer demosaicking," IEEE Trans. Image Process, vol. 20, July 2011, pp. 1885-1894. (IEEE Xplore citation) This paper expands on the work in the ICIP 2006 paper above, optimizing the LSLCD performance-complexity tradeoff and comparing with several state-of-the-art techniques. The proposed method had the best quality-complexity tradeoff of the methods compared. There is a web page for the paper with more complete results and software to reproduce the results.

• G. Jeon and E. Dubois, “Demosaicking of noisy Bayer-sampled color images with least-squares luma-chroma demultiplexing and noise level estimation,” IEEE Trans. Image Process., vol. 22, Jan. 2013, pp. 146-156. doi: 10.1109/TIP.2012.2214041. This paper extends the work of Leung et al. above by adding a step of noise-level estimation, following by adaptation of the luma-chroma demultiplexing algorithm to the estimated noise level. There is a web pager for the paper with additional results and software.

• P. Hao, Y. Li, Z. Lin and E. Dubois, "A geometric method for optimal design of color filter arrays," IEEE Trans. Image Process., vol. 20, pp. 709-722, March 2011. (IEEE Xplore citation) Web page for the paper with more results and software.