curvature filters are efficient solvers for variational models
Posted: 2015-07-23T23:17:54-07:00
Curvature regularized variational models are usually difficult to solve. Traditionally, the solvers are either based on diffusion scheme (gradient decent method) or based on Euler Lagrange Equation. The former suffers from numerical issues while the later one usually has a very complex form.
Fortunately, there are some filters that can efficiently minimize the Gaussian or mean curvature without computing any of them. The source code and theoretical explanation can be found at https://github.com/YuanhaoGong/CurvatureFilter
These filters can be used in a large range of image processing problems, such as denoising, blind deconvolution (debluring), segmentation, dehazing, enhancement, etc, because they can solve arbitrary imaging models as long as the imaging model can be evaluated (black box).
Wish you like these filters! Cheers!
Fortunately, there are some filters that can efficiently minimize the Gaussian or mean curvature without computing any of them. The source code and theoretical explanation can be found at https://github.com/YuanhaoGong/CurvatureFilter
These filters can be used in a large range of image processing problems, such as denoising, blind deconvolution (debluring), segmentation, dehazing, enhancement, etc, because they can solve arbitrary imaging models as long as the imaging model can be evaluated (black box).
Wish you like these filters! Cheers!