基于偏移差值的结构化稀疏图像修复算法

Structured sparse image inpainting algorithm based on offset difference

针对基于匹配的图像修复算法中存在线性结构和重复区域不能良好保持结构连贯性的问题,提出基于偏移差值的结构化稀疏图像修复算法。首先定义一个新的特征偏移差值,结合图像颜色特征构造样本块间相似度项,并利用该相似度项寻找最佳匹配块;然后将偏移差值信息和颜色信息共同作为稀疏约束项来构造约束方程,以提高稀疏表示的准确性,从而保证邻域信息一致性。实验结果表明,文中方法对线性结构和重复区域的破损修复能获得更好的效果。

Aiming at the problem that the matching-based image restoration alg orithm can not maintain the structural consistency of the linear structure and the repeated region,a struct ured sparse image restoration algorithm based on the offset difference is proposed.Firstly,a new feature offset difference is defined, and the similarity between the sample blocks is constructed by combining with th e image color feature,and the similarity item is used to find the best matching block;then the offset differe nce information and the color information are collectively used as sparseness.The constraint term is used to construct the constraint equation to improve the accuracy of the sparse representation,thus ensuring the consistency of the neighborhood information. The experimental results show that the proposed method can achieve better result s for the damage repair of linear structuresand repeat regions.