基于图像边缘检测的扭曲文档矫正

Correction of distorted documents based on image edge detection

扭曲的文档图像会干扰文档图像的光学字符识别(Optical Character Recognition,OCR).为了对扭曲形变的文档图像进行矫正,提高扭曲文档识别的正确率,基于目标检测与分割的网络,提出文档图像的边缘检测方法,使用贝塞尔(Bezier)曲线拟合文档图像的边缘曲线,通过目标检测的算法回归Bezier曲线的控制点.将文档图像的边缘检测转化为边缘曲线Bezier控制点的回归,使用文档的边缘点计算扭曲文档矫正后的矩形模板,然后将文档图像通过薄板样条插值(Thin Plate Spline,TPS)算法重映射到矩形模板中,完成文档的矫正.实验结果表明,提出的矫正方法能够对扭曲文档进行精确的边缘提取,和其他算法相比,经该算法矫正后的文档图像,其OCR的正确率有较大的提升.

Distorted document images interfere with optical character recognition(OCR)of document images.To correct distorted document images and improve the correct rate of distorted document OCR recognition,this paper proposes an edge detection method for document images based on the object detection and segmentation network,uses Bezier curves to fit the edge curves of document images,and returns the control points of Bezier curves through the object detection algorithm.Convert the edge detection of the document image into the regression of Bezier control points of the edge curve,use the edge points of the document to calculate the rectified rectangular template of the distorted document,and then remap the document image to the rectangular template through the thin plate spline algorithm to complete the correction of the document.Experimental results show that the proposed correction method accurately extracts the edges of distorted documents.Compared with other algorithms,the corrected document image has a greater improvement in the accuracy of OCR.