结合自适应梯度分类匹配的改进Criminisi图像修复算法

Improved Criminisi Algorithm Based on Adaptive Gradient Classification and Matching

传统Criminisi修复算法优先权不合理且不能自适应计算,易导致错误填充和纹理延伸问题,而在更新已知区域全局匹配,不仅提高了时间复杂度也降低了匹配准度,而对搜索空间添加阈值限制则会降低修复质量.针对以上问题,提出一种自适应梯度分类匹配的改进Criminisi修复算法.所提方法首先将图像初始已知区域像素按梯度直方图自适应地划分为平滑、纹理和边缘3种类型;其次结合自适应块分类优先权函数来克服纹理延伸问题;最后由梯度估计待修补块类型和自适应块大小函数来保证待修补块只在初始已知区域对应大小类型块中匹配来提高匹配效率和避免使用新增已知区域.实验表明所提方法较好克服了纹理延伸和时间复杂度高等问题并提高了修复质量.

In conventional Criminisi inpainting algorithm,the unreasonable priority and non-adaptive computing lead to error filling and texture extension.The global matches in update known regions not only increase time complexity but also affect final restoration quality while limiting search space by threshold decreases image inpainting quality.To address these problems,an improved Criminisi inpainting algorithm based on adaptive gradient classification and matching was proposed.Firstly pixels in image initial known regions were adaptively classified into three types such as smooth,texture and edge by gradient histogram;Secondly an adaptive patch classification priority function was used to overcome texture extension;finally the estimated patch type and adaptive size function by gradient were used to guarantee high matching efficiency and avoid update known areas by only matching proper patch type size in initial known regions.The experiments show the proposed method can avoid texture extension,decrease time complexity and improve image inpainting quality.