基于梯度变换与最优似然法则的图像修复算法
An Image Inpainting Method Based on Gradient Transformation Coupled with Optimal Likelihood Rule

为了解决当前图像修复算法在待修复图像纹理结构较为丰富时易产生模糊效应以及块效应的问题，提出了一种基于梯度变换与最优似然法则的图像修复算法.首先，利用像素点对应的邻域方向特征来构造置信度，用以形成优先权因子.通过优先权因子对待修补块的优先级进行度量，从而确定最优修补块；然后，根据像素点的梯度变换，构造修补块尺寸选择模型，对修补块的尺寸进行自适应调整；最后，利用修补块与匹配块的内积关系、距离关系，分别构造余弦度量模型、相似度量模型，从而建立最优似然法则，从源区域中搜索最优匹配块，对待修复块进行填充修复.实验结果显示，与其他图像修复算法相比，本文算法具备更高的修复质量，能有效克服阶梯效应以及模糊效应.
In order to solve such defects in current image inpainting algorithms as blurring effect and blocking artifact induced by taking the repaired block with fixed size as template to search the optimal matching block for image inpainting, an image inpainting method based on gradient transformation coupled with optimal likelihood rule is proposed in this paper. Firstly, the confidence is constructed by using the neighborhood direction feature of pixel points to form the priority factor. And the optimal patch block is determined by using the priority factor to measure the priority of the patch blocks. Then, a repair block size selection model is constructed based on the gradient transformation of the pixel points to adaptively adjust the size of the patch block. Finally, the cosine metric model and similarity measurement model are constructed, respectively, based on the inner product relation and the distance relation between patches and matched blocks to establish the optimal likelihood rule for searching the best matching block from the source region and filling and repairing the repaired blocks. The results of an experiment have shown that compared with the current image inpainting algorithms, this algorithm has better repair quality which can effectively overcome the staircase effect and the blur effect