Abstract: There exits the discontinuous edge and low efficiency problem when the image inpainting algorithm is used to deal with the stained license plate image, a new improved algorithm based on the combination of texture blocks and gradient feature for image inpainting is proposed. The new algorithm is improved with two main aspects. Aiming at the optimal matching block, the gradient calculation is added to optimize the selection and pre-editing for the matching block, so as to be integrated with the original image. A modified scheme to update the filling-front is adopted to improve the inpainting efficiency. Experimental results indicate that the proposed improved algorithm has a better computation efficiency, and has a better edge continuity in visual perception.

Key words: image inpainting, gradient calculation, pre-editing, filling-front update, computational efficiency

摘要： 针对目前图像修复算法在处理污损车牌图像时存在边缘修复不连续和效率不高问题，提出采用纹理块与梯度特征结合的改进的算法。新算法从两个方面进行改进：针对最优匹配块，加入梯度计算，优化匹配块的选择及预编辑，提高与环境的融合效果；采用了一种更新填充前端的修改方案，提高修复效率。测试结果表明，所提出的算法具有更好的计算效率，视觉感知方面具有更好的边缘连续性。

关键词: 图像修复, 梯度计算, 预编辑, 更新填充前端, 计算效率