A New Oriented-Diffusion Image Inpainting Framework for Striped Texture Images

Yong Zhu, Gui Wang, Zhike Han

Abstract: Modeling texture image is a challenge in image processing. The classical PDE model proposed for digital image inpainting can not be directly applied to striped texture image inpainting, since variation analysis of the PDE model shows that it may lead to trivial solution. In this paper, A novel PDE framework is proposed for striped texture image inpainting. The major idea is to employ the orientation field to act as the constraint of the diffusion directions. Once the orientation field can be correctly estimated, the gray level information can be propagated into the inpainting domain along a local fix orientation, which characterizes the orientation of the stripes. This paper is also an attempt to proposing an inpainting model which explicitly encoding spatial or geometrical correlations which mean orientation field in our models, especially for striped texture image or texture image with similar patterns. Numerical experimental results show that the proposed models are superior to common PDE models when inpainting striped texture images.