Equalized Net Diffusion (END) in Image Denoising

Abstract: This article is concerned with the method of diffusion modulation, which can be applied for various conventional partial differential equation (PDE)-based restoration models in order to effectively restore not only fine structures but also slow transitions. We in particular introduce the equalized net diffusion (END) which tries to equalize the anisotropic diffusion over a wide range of image content. Although the new reformulated models incorporating END are highly nonlinear, they can be simulated efficiently by adopting linearized stable numerical procedures. The END-incorporated models have outperformed over the basic (conventional) PDE-based models in both quality and efficiency. Key-Words: Image restoration, variational approach, diffusion modulation, equalized net diffusion, PDE-based image denoising.