Perceptual Color Image Smoothing via a New Region-Based PDE Scheme

Baptiste Magnier (1) , Philippe Montesinos (1) , Daniel Diep (1)

1 LGI2P - Laboratoire de Génie Informatique et Ingénierie de Production

In this paper, we present a new color image regularization method using a rotating smoothing filter. This approach combines a pixel classification method, which roughly determines if a pixel belongs to a ho- mogenous region or an edge with an anisotropic perceptual edge detector capable of computing two precise diffusion directions. Using a now classical formulation, image regularization is here treated as a variational model, where successive iterations of associated PDE (Partial Differential Equation) are equivalent to a diffusion process. Our model uses two kinds of diffusion: isotropic and anisotropic diffusion. Anisotropic diffusion is accurately controlled near edges and corners, while isotropic diffusion is applied to smooth re- gions either homogeneous or corrupted by noise. A comparison of our approach with other regularization methods applied on real images demonstrate that our model is able to efficiently restore images as well as handle diffusion, and at the same time preserve edges and corners well.

Mots clés: half Gaussian kernels; Anisotropic diffusion; edge detection; steerable filters; half Gaussian kernels.

Domaines: Traitement des images [eess.IV]

Baptiste Magnier, Philippe Montesinos, Daniel Diep. Perceptual Color Image Smoothing via a New Region-Based PDE Scheme. Electronic Letters on Computer Vision and Image Analysis, 2013, 1, pp.17--32. ⟨hal-00807751⟩