【作者】 田艳艳；

【导师】 彭进业；

【作者基本信息】 西北大学 ， 信号与信息处理， 2008， 硕士

【摘要】 数字图像修补技术在污损图像修补、目标移除、图像压缩等图像处理领域扮演着很重要的角色。本文主要研究基于偏微分方程的图像修补方法。在污损区初始化方面,本文考虑到污损区域周围信息与缺失信息的相关性,提出“相关随机初始化”方案。相较于其它初始化方案,该方案能够有效改善修补效果并提高修补效率。在修补模型方面,本文在整体变分模型的基础上,分析了图像的梯度和水平集曲率对传导系数的影响,提出G-C修补模型,使得修补效果更加自然。与此同时,综合考虑沿等照度线切向与法向的扩散,将G-C模型和BSCB模型进行组合,提出统一修补模型。实验结果证明,相比于G-C模型,统一模型的修补效率有很大提高。更多还原

【Abstract】 Digital image inpainting plays an important role in image processing, including inpainting destroyed images, disocclusion, image compression and so on. The paper focuses on research of image inpainting models based on PDE.About the initialization of destroyed area, we take the relativity of the destroyed area and the natural area into consideration, and then the "related random initialization" is proposed. The experimental results show that compare with previous schemes, the method can make the inpainting outputs more natural and improve the inpainting efficiency as well.About the inpainting models, we mainly observe the two important characters of images, the gradient and the curvature of the isophotes, and introduce the G-C model based on total variation. The model can restore the information of the destroyed area naturally and connect the broken level sets smoothly. Meanwhile, we combine the diffusion along the tangent and normal direction, and then present a combined model based on G-C model and BSCB model. The results show that the novel model performs better than G-C model in inpainting efficiency.更多还原