摘要
目前的名片OCR系统普遍采用Hough变换法和K最近邻簇法来检测和矫正名片图像倾斜角度，但是这些算法无法矫正名片的透视图像，该文从射影几何角度提出了基于射影矩阵变换的算法：利用名片图像固有的边缘直线信息，将名片与名片图像之间的透视变换矫正为仿射变换，再将仿射变换矫正为相似变换，此相似变换实际上是对名片进行旋转和全局缩放的结果，消除其中的旋转因素，就得到一个与名片仅差一个缩放因子的正交名片图像，以便于后续的字符处理。
Most of OCR systems detect and rectify tilt angle by Hough transform or KNN generally. However these algorithms are ineffective in perspective Image Rectification,This paper analyzes the defect of these algorithms and proposes a new＇ correc- tion algorithm based on projective matrix transformation. The developed algorithm is as follows： By using the given prior knowl- edge of Business Card, the perspective transformation between Business Card and its image is firstly converted to affme transfor- mation. Then the obtained affine transformation is converted to similarity transformation, which contains a rotation and global scaling. After removing the effect of rotation factor, a Orthogonal Business Card image is obtained, which only has a scale factor compared with real Business Card and is convenient for following character processing.
作者
唐维
任国强
TANG Wei, REN Guo-qiang（1. KaiLi University, Information ＆ Network Center, Kaili 556011,China; 2. Shandong University of Sci-Tech, Department of Electric Information, Jinan 266510,China）
出处
《电脑知识与技术》
2013年第9期5711-5715,共5页
Computer Knowledge and Technology
关键词
名片
透视图像
透视变换
仿射变换
相似变换
business card
perspective images
perspective transform
affine transform
similarity transform
