基于字典学习的古建筑图像修复研究  

Study on inpainting of drawings of ancient buildings based on dictionary learning

作　　者：杨茹[1] 秦振涛[1] 杨武年[2] 

机构地区：[1]攀枝花学院数学与计算机学院,四川攀枝花617000 [2]成都理工大学地学空间信息技术国土资源部重点实验室,四川成都610059

出　　处：《电子技术应用》2016年第12期51-53,57,共4页Application of Electronic Technique

基　　金：国家自然科学基金资助项目(41372340);高等学校博士学科点专项科研基金(20105122110006);国土资源部地学空间信息技术重点实验室开放基金资助项目(KLGSIT2014-03)

摘　　要：古代建筑图纸是我国重要的民族瑰宝,亟待利用现代计算机技术对古代建筑图纸进行信息化和修复。提出了一种新的基于字典学习的古建筑图像修复模型,通过K-svd算法进行字典学习,在稀疏域利用已知像素信息填充缺损像素,从而实现对古建筑图像的修复及噪声的滤除。实验表明,该算法能较好地修复古建筑图像,降低图像的均方误差,在实际应用中具有良好的可行性和应用前景。Drawings of ancient buildings are tenures of traditional Chinese culture, and need to be informatization and inpainting urgently. A novel model of inpainting the drawings of ancient buildings is proposed, which carry on the dictionary learning by Ksvd algorithm, and filling missing pixels by known pixels in sparse domain, so as to implement the inpainting of drawings of ancient buildings and remove impulsive noise. The experimental results show that the algorithm can inpaint the image more effectively and decrease the RMSE, this method has better performance than other dictionary learning algorithm, and has good application potential and good application prospects.

关 键 词：稀疏表示 古代建筑图纸 字典学习 图像修复 

分 类 号：TN911[电子电信—通信与信息系统]