An interactive image inpainting algorithm based on Markov Random Field

A novel approach to image inpainting is introduced in this paper. The novelty lies in the combination of pixel-diffusing technique and a user interaction mechanism. This combination takes both local geometrical information via the pixel-diffusing technique and global structure information via the user interaction to improve the image inpainting quality. The user interaction mechanism manually specifies important missing structure information by drawing some curves from the known to the unknown regions and synthesizes image structure along these user-specified curves in the unknown region using structure information selected around the curves in the known region. The pixel-diffusing technique builds on the Markov Random Field (MRF) model to exploit the image contextual knowledge. The interactive inpainting algorithm fills in the remaining unknown regions based on the MRF model. The experiment results show this interactive algorithm is reasonable and efficient. ©2009 IEEE.