An image inpainting algorithm based on local properties

作者：Fan Guan Peng^*; He Hong Jie; Chen Fan; Zhai Dong Hai; Renqingnuobu

来源：Journal of Optoelectronics - Laser, 2012, 23(12): 2410-2417.

• Block effects
• Compensation coefficients
• Damaged region
• Data terms
• False matches
• Gradient informations
• Image Inpainting
• Image Structures
• Inpainting
• Local gradients
• Local property
• Local search spaces
• Local similarity
• Search spaces
• Structure characteristic

In the current exemplar-based image inpainting algorithms, there are some shortcomings, such as the unreliability of the filling order, the improper template window of sample block and the low efficiency of inpainting, etc. To address these problems, an image inpainting algorithm is proposed based on local properties in this paper. Combining the local gradient features and the local similarity characteristics of image, the proposed algorithm increases the compensation coefficient and weight of the data term to ensure the connectivity of image structure. The size of template window of sample block is chosen by the gradient information of image to reduce the error derivative and block effects. The confidence term is used to adaptively select the size of local search space to improve efficiency of inpainting and reduce false match. Experimental results show that the proposed algorithm can maintain good texture and structure characteristics of the damaged region.

• 出版日期2012
• 单位西南交通大学