Light fields denoising using 4D anisotropic diffusion
 

In this paper, we present a novel light field denoising algorithm using a vector-valued regularization operating in the 4D ray space. More precisely, the method performs a PDE-based anisotropic diffusion along directions defined by local structures in the 4D ray space. It does not require prior estimation of disparity maps. The local structures in the 4D light field are extracted using a 4D tensor structure. The paper then describes the strategy retained for setting the diffusion tensor parameters for the targeted denoising application. It then analyzes the influence of the model parameters on the denoising performance. Experimental results show that the proposed denoising algorithm performs well compared to state of the art methods while keeping tractable complexity.

In this work, real light fields catured by Lytro Illum cameras from the EPFL dataset are being used. The Illum light fields are decoded and extracted by Dansereau's Matlab Light Field Toolbox.

AnkylosaurusDiplodocus

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 33.14 dB

Bike

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 30.10 dB

ColorChart

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 30.82 dB

DangerDeMort

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 31.20 dB

Desktop

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 31.20 dB

Flowers

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 30.54 dB

FountainVincent

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 29.47 dB

Friends

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 31.43 dB

IsoChart

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 30.37 dB

Magnets

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 33.42 dB

StonePillarsOutside

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 30.31 dB

Vespa

Noised, σ = 50/255, PSNR 14.15 dB	Diffusion over 30 iterations at center view	Denoised, PSNR 32.26 dB

Side by side comparison with state of the art method

Bike

Noised, σ = 50/255, PSNR 14.15 dB, Denoised with LFBM5D, PSNR 29.90 dB	Denoised with the proposed method, PSNR 30.10 dB

Magnets

Noised, σ = 50/255, PSNR 14.15 dB, Denoised with LFBM5D, PSNR 33.26 dB	Denoised with the proposed method, PSNR 33.42 dB