Minimization of the Image Surface Curvature via Image Reconstruction Using Discrete Mean and Gaussian Curvature
摘要
The applied curvature regularities offer excellent priors in the continuous edges to various computer vision and image processing applications. First-order optimality for these types of problems requires the use of high-order parametric differential equations due to their non-convex, non-smooth, overall nonlinear characteristics. This is the reason why the computation of numbers is highly demanding. This work computes the discrete mean curving and Gaussian curvature by utilizing the shape of the localized 3 × 3 stencil, which is a fundamental concept in differential geometry. Using multipliers in alternating directions, the problem of curvatures over the picture surface can be efficiently solved by lowering some functions, which is a sort of weighted image surface reduction problem. The key benefit of utilizing a Gaussian curvature regularizer is that it can maintain sharpness, edges, and corners very well while providing an accurate curvature estimation. Numerical tests on picture recovery and painting are done in order to show the effectiveness and quality of the offered curvature-based model over current variational approaches. The MATLAB programming language with a version of R2017b or higher is to be used in this work.