基于学习字典的图像修复算法  

Image inpainting algorithm based on learned dictionary

作　　者：李民[1,2] 李世华[1] 乐翔[3] 李小文[1] 罗环敏[1] 

机构地区：[1]电子科技大学地表空间信息技术研究所,成都611731 [2]桂林空军学院科研部,桂林541003 [3]电子工程学院,成都611731

出　　处：《仪器仪表学报》2011年第9期2041-2048,共8页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(No.50875070)资助项目

摘　　要：多数图像修复方法主要是利用输入图像中的有效信息来填充待修复区域,可用先验信息有限,自适应性较差。研究提出一种新的基于学习字典的图像修复框架,核心思路是通过大量样本图像和输入图像的有效数据训练学习字典,建立样本图像特征块与原始数据块之间的稀疏关联,并将这种关联作为先验知识来指导图像的修复。该方法既充分利用了样例图像的先验知识,又考虑了待修复图像本身的信息,提高了算法的自适应性。通过对自然图像进行大﹑小范围图像修复和文字去除实验,文中方法均取得较好的修复效果。Most recent image inpainting methods only use valid information found in input image as the clue to fill the inpainting region. These methods usually have the defects of insufficient prior information and relatively poor adaptivity. A novel learned dictionary based image inpainting framework is presented. The key idea is to build a sparse relationship between raw image patches and their corresponding feature patches, then use this relationship as the priori to guide the inpainting. Our method not only uses the valid information of the input image itself, but also utilizes the prior information of the sample images to improve the adaptivity. Large and small region inpainting experiments and text removing experiments on nature images show the good performance of our method.

关 键 词：图像修复 稀疏表示 学习字典 L1范数 在线字典学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]