Total Variation Grayscale and Color Image Denoising

The Rudin-Osher-Fatemi total variation (TV) denoising technique poses the problem of denoising as a minimization,
   Min_u int |grad u|+ (lambda/2) int (f-u)^2
where f is the noisy image, lambda is a nonnegative parameter, and u is the denoised image we seek.

u = tvdenoise(f,lambda) denoises the input image f using Chambolle's method [1]. The smaller the parameter lambda, the stronger the denoising.

If f is a color image (or any array where size(f,3) > 1), the vectorial generalization of the TV model is used and solved by a generalization of Chambolle's method [2]. The screenshot shows an example of tvdenoise applied to a noisy color image.

References
[1] A. Chambolle, "An Algorithm for Total Variation Minimization and Applications," J. Math. Imaging and Vision 20 (1-2): 89-97, 2004.
[2] X. Bresson and T.F. Chan, "Fast Minimization of the Vectorial Total Variation Norm and Applications to Color Image Processing", CAM Report 07-25.

Uses a faster minimization method and now supports color image denoising.