一种有效的大区域图像修复算法

摘要

For the optimisation of image inpainting,ordinary non-texture image inpainting or texture synthesis algorithm can’t perform very well in repairing the natural image with large area as the inpainting area.Traditional CPT algorithm presents a very good thought to tackle the natural image inpainting,it combines the texture synthesis with non-texture image inpainting,and improves the effectiveness of inpainting process by using the iterative algorithm and carry out block matching each time on the point with highest priority,but the deficiency cannot be fully excluded.This paper proposes an effective large-area image inpainting algorithm by analysing the problems existed in CPT algorithm and making the enrichment and improvement under its framework.In addition it also gives a novel gradient-priority calculation method.This algorithm turns out to be effective by experiment in improving the quality of large-area images inpainting.%关于图像修复优化，普通的非纹理图像修复或纹理综合算法都不能对修复区域为大区域的自然图像取得很好的修复结果，传统的 CPT 算法提出了一种很好的解决自然图像修复思路，结合纹理综合和非纹理修复的方法，采用迭代算法，每次对优先权最高的一点进行块匹配，以此来提高修复过程的有效性，但也存在不足。通过对 CPT 算法存在的问题进行分析，并在其框架下进行充实和改进，提出一种有效的大区域图像修复算法，并给出一种全新的梯度优先的计算方法。实验证明该方法是有效的，提高了大区域图像修复效果。