视觉文档图像的几何校正  

Geometric rectification for camera-captured document images

作　　者：田学东[1] 马兴杰[1] 韩磊[2] 刘海博[1] 

机构地区：[1]河北大学数学与计算机学院,河北保定071002 [2]河北省气象技术装备中心,石家庄050021

出　　处：《计算机应用》2007年第12期3045-3047,共3页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60772073)

摘　　要：在以数码相机等数字设备拍摄文档资料时,所拍摄的图像经常会产生各种各样的几何变形,这种变形可能会导致识别软件中的版面分析和切分算法失效,从而使文档图像无法被识别。为了使普通的识别软件能够对数码相机等拍摄的文档图像进行识别,有必要对其进行几何校正。根据几何变形产生的原因对其进行了分类,并针对不同种类的变形提出了相应的校正算法。实验结果证明该分类方法和相应校正算法都有较好的效果。The document images filmed by digital devices such as digital cameras are always coupled with distortions, which will result in a failure of the layout analysis, segmentation and recognition. In order to enable ordinary recognition software to recognize the digital document images, it is necessary to carry out their geometric rectification. A new identification technique was proposed to distinguish three types of image distortions and then the restoration of camera images was carried out respectively. Experimental results show that the proposed classification method and restoration techniques are effective.

关 键 词：OCR 视觉文档图像 几何变形 透视变形 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]