Deep Joint Design of Color Filter Arrays and Demosaicing

This is the supplementary material for the paper Deep Joint Design of Color Filter Arrays and Demosaicing, presented in Eurographics 2018 and published in a special issue of Computer Graphics Forum. We show additional reconstructions, comparing our method for noise-free and noisy scenarios.

The images below are a random sample from the moiré and vdp datasets [GCPD16]. Best PSNR values shown in bold.

Important: These materials are best viewed in a web browser that is not Internet Explorer (IE), since it cannot resize images properly.

Please click on any image to see them in a larger size, then use the arrow keys from your keyboard for navigating between the images.

Noise-free
Noisy (σ=20)

Here, we compare Our 4x4 noise-free against the following state-of-the-art techniques:

[Cha16] Chakrabarti: Learning sensor multiplexing design through back-propagation. CoRR abs/1605.07078 (2016).
[MBP*09] Mairal et al.: Non-local sparse models for image restoration. In ICCV (2009), pp. 2272-2279.
[HST*14] Heide et al.: Flexisp: A flexible camera image processing framework. ACM TOG 33, 6 (2014).
[GCPD16] Gharbi et al.: Deep Joint Demosaicking and Denoising. ACM TOG 35, 6 (2016), 191:1-191:12.

moiré

vdp

Here, we compare Our 4x4 noise against the following state-of-the-art techniques:

[Cha16] Chakrabarti: Learning sensor multiplexing design through back-propagation. CoRR abs/1605.07078 (2016).
[CM12] Condat and Mosaddegh: Joint demosaicking and denoisingby total variation minimization. In ICIP (2012), pp. 2781-2784.
[GCPD16] Gharbi et al.: Deep Joint Demosaicking and Denoising. ACM TOG 35, 6 (2016), 191:1-191:12.