A Generalized Vector-Valued Total Variation Algorithm

Total Variation algorithm for denoising/deconvolving grayscale/color (vector) images

Description

Compute the minimum of the generalized TV functional

T = || K*U - S ||^p + lambda*|| sqrt( (Dx(U))^2 + (Dy(U))^2 ) ||^q

for grayscale / color (vector) images using the IRN [1,2] algorithm, where

S: Input image
lambda: regularization parameter
K: linear operator
U: Output image

References

[1] P. Rodriguez, B. Wohlberg, "Efficient Minimization Method for a Generalized Total Variation Functional"
IEEE Transactions on Image Processing, 2009, 18:2(322-332)

[2] P. Rodriguez, B. Wohlberg, "A Generalized Vector-Valued Total Variation Algorithm" submmited to ICIP'09 (http://www.icip2009.org/)

• MATLAB release: MATLAB 7.3 (R2006b)

[Vijay]

Algorithm seems to work really well for SNR < 15 where SNR = mean/std(noise). For higher SNR it seems to smooth out features. Explaination may be useful.