基于虚拟成像系统的文档图像矫正

Document Image Correction Based on Virtual Imaging System

由于相机成像系统中相机、成像平面、文档平面三者位姿角度的原因,会导致相机获取的文档图像为透视投影而不是正视图,对后期利用计算机视觉技术对文字的自动化提取与识别等处理造成消极影响从而降低文字提取与识别的准确率;对相机成像系统进行分析,影响相机成像的因素是数学模型下的相机的成像坐标系与世界坐标系的相互转换问题,所以利用了相机坐标系经旋转矩阵转换的技术,提出了一种通过旋转矩阵将真实成像相机坐标系转换为正视图的虚拟相机坐标系,并进行图像像素点的重映射,以实现对文档图像的位姿校正;采用双目相机进行文档拍摄实验,结果表明该方法过程简单易行、可行性高、能有效地对文档图像进行矫正,获得文档图像的正视图。

Due to the posture of camera,imaging plane and document plane in the camera imaging system,the document image obtained by the camera will be perspective projection instead of front view,which will have a negative impact on the later processing of text automatic extraction and recognition by computer vision technology,the accuracy of text extraction and recognition is reduced.The camera imaging system is analyzed,the conversion between the camera imaging coordinate system and world coordinate system is a factor affecting the image.The principle of camera imaging system and rotation matrix is used to propose a virtual camera coordinate system which converts the real camera coordinate system into the front view through the rotation matrix,and the image pixels are remapped to realize the pose correction of the document image.In this paper,the document experiment is captured by using a binocular camera,the results show that the method is simple and feasible for the correction of document images,and obtains the front view of document images.