基于球形收敛和结构一致性的图像修复算法

Image Inpainting Algorithm Based on Spherical Convergence and Structure Consistency

为获得稳定的填充顺序及合适的匹配准则,提出基于球形收敛和结构一致性的样本块图像修复算法。在填充顺序方面,一方面将球形收敛规则引入到优先权准则的定义中,在保持优先填充结构部分的同时保证纹理和平滑部分的合理延伸;另一方面构造基于斯特林理论的置信度项更新准则,以降低置信度的快速衰减。在匹配准则方面,引入结构一致性以寻找更加合适的样本块进行填充,从而减轻误匹配和误差累积现象。实验结果表明,本算法较现有图像修复算法能获得更加稳定的修复次序,使补全后图像的结构连贯且纹理部分过渡自然,同时有效降低了误差累积效应,获得较好的修复效果。

To obtain the robust filling order and suitable matching criterion,an image inpainting algorithm based on spherical convergence and structure consistency was proposed.In terms of filling order,on the one hand,the spherical convergence rule was introduced into the definition of the priority criterion to ensure the reasonable extension of the texture and smooth part while maintaining the priority filling of the structural part.On the other hand,a confidence item update criterion based on Stirling theory was constructed to avoid rapid decay of confidence value.In terms of matching criterion,the structural consistency was applied to find a more suitable exemplar for filling to reduce false matching and error accumulation.The experimental results show that compared with the existing image inpainting algorithms,the proposed method can produce a more stable inpainting order,maintain structure coherence and texture consistency of the completed image,effectively reduce error accumulation effect,and obtain a better restoration effect.