基线自适应透视变换的文本行矫正  

Baseline adaptive perspective transformation for text-line image correction

作　　者：张梦林 杨淑莹 ZHANG Mengin;YANG Shuying(School of Computer Science and Engineering,Tianjin University of Technology,Tianjin 300384,China;Key Laboratory of Computer Vision and System,Ministry of Education,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学计算机科学与工程学院,天津300384 [2]天津理工大学计算机视觉与系统教育部重点实验室,天津300384

出　　处：《天津理工大学学报》2024年第4期76-82,共7页Journal of Tianjin University of Technology

基　　金：天津市教育科学规划院教学成果奖重点培育项目(PYGJ-015);天津理工大学校级重点教学基金(ZD20-04)。

摘　　要：相机拍摄的文档图像通常存在弯折和透视形变,这将导致由图像提取的文本行弯曲和文字的大小不一致。提出基线自适应透视变换来进行文本行矫正。该方法使用Bezier曲线拟合文本行中心和上、下边界基线,在文本行拉直矫正中加入了横向矫正效果。提出的方法将需要矫正的文本行片段模拟为倾斜平面,当文本行片段高边方向与文档旋转轴向角度为45°时,未经过透视形变与经过透视形变的文本行片段高度比与宽度比的比值相同。根据片段高度与文本行平均高度比值进行宽度变化并计算透视变换矩阵,矫正其中存在的透视形变。对实际拍摄的文档图像提取的文本行进行人工检查,将没有完成的文本行拉直矫正,以及矫正后有字体较大错误形变的文本行图像作为矫正失败的文本行图像,文本行矫正成功的概率约为98.08%。The document image taken by camera usually has bending and perspective deformation,which will lead to the bending of the text line extracted from the image and the inconsistent size of the text.The baseline adaptive perspective transformation is proposed to correct the text line.Bezier curve is used to fit the center,upper and lower boundary baselines,and the horizontal correction effect is added to the text line straightening method.The proposed method simulates the text line segment to an inclined plane.When the angle between the segment high side direction and the rotation axis of the document is 45°,the height ratio and the width ratio between the original text line segment and the text line segment with perspective deformation is the same.According to the ratio of the segment height to the average height of the text line,the segment width after correction is dynamically determined,and the perspective transformation matrix is calculated to correct the perspective deformation.The text lines extracted from the actual document image is checked manually,and take the text line images that has not completed the text line straightening and has large text deformation error after correction as the failed cases.The probability of successful text line correction is about 98.08%.

关 键 词：文档矫正 文本行拉直 透视变换 基线估计 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]