基于置信特征与结构相似度约束的图像修复算法

Image Inpainting Algorithm Based on Confidence Feature Coupling Structure Similarity Constraint

当前图像修复算法主要通过对像素值的差异进行度量来完成图像修复,当待修复图像中破损区域较大时,该方法修复的图像中会出现较为明显的块效应以及纹理不连续效应,导致算法的修复效果不佳.对此,本文提出了基于置信特征耦合结构相似度约束的图像修复算法.引入优先权函数,对破损区域中待修复像素的优先权值进行计算,以确定待修复像素.利用待修复像素的梯度模值来构造梯度变化约束机制,对修复块的模板大小进行调整,以提高图像的修复质量.通过像素点的置信度构造置信特征因子,并利用置信特征因子建立了搜索空间适配法则,明确了最优匹配块的搜索范围.引入结构相似度模型,对修复块与匹配块之间的结构相似度进行约束,以选取最优匹配块,利用最优匹配块对修复块进行扩散填充,从而完成图像的修复.仿真实验表明:与当前图像修复技术相比,本文方法具有更好的图像复原质量,有效降低了块效应以及不连续效应.

At present, the image inpainting algorithm mainly completes the image restoration by measuring the difference of the pixel values, when the damaged region in the image to be repaired is larger, there will be obvious block effect and discontinuous texture effect in the restored image, which will lead to a poor repair effect. An research of image inpainting algorithm based on confidence feature coupling structure similarity constraint was proposed in this paper. The priority function is introduced to calculate the priority weights of the pixels to be repaired in the damaged area to determine the pixels to be repaired. A gradient constrained model is constructed by using the gradient modulus of the repaired pixels, and the template size of the repaired blocks is adaptively adjusted to improve the inpainting quality. The confidence feature factor is constructed by the confidence of the pixels, and the search space adaptation rule is established by using the confidence feature factor. The search range of the optimal matching block is defined. The introduction of structural similarity index measurement system model, to measure the repair block and matching the structural similarity between the blocks, to select the optimal matching block by block matching to repair block diffusion filling, thus completing the image restoration. Simulation results show that the proposed method does not have block effect and discontinuity, and has better visual effects.