tvreg: Variational Imaging Methods

The tvreg package performs total variation (TV) regularized image denoising, deconvolution, and inpainting. Three different noise models are supported: Gaussian (L2), Laplace (L1), and Poisson. The implementation solves the general TV restoration problem

to perform denoising, deconvolution, and inpainting as special cases. It is efficiently solved using the recent split Bregman method. Also included is an efficient implementation of Chan-Vese two-phase segmentation. All functions support grayscale, color, and arbitrary multichannel images.

Please see the included documentation file tvreg.pdf for details.

Get Started Quickly

1. Install the FFTW3 library. Windows users can download precompiled DLL files from http://www.fftw.org/install/windows.html.
2. Compile the programs with GCC using make -f makefile.gcc or Microsoft Visual C++ using nmake -f makefile.vc. See section 7 of tvreg.pdf for help.
3. Try the demos

tvdenoise_demo: Denoising demo

tvdeconv_demo: Deconvolution demo

tvinpaint_demo: Inpainting demo

chanvese_demo: Segmentation demo

Get Started Quickly in MATLAB

Compiling is not required to use tvreg in MATLAB. Try the demos

tvdenoise_demo: Denoising demo

tvdeconv_demo: Deconvolution demo

tvinpaint_demo: Inpainting demo

chanvese_demo: Segmentation demo

For improved performance, run the included script complex_mex.m to compile the main computation routines as MEX functions. This requires that FFTW3 is installed, please see section 7.3 of the documentation.