中文摘要：

OCR在识别数码相机获取的文档图像时，常常会因为图像的扭曲而导致识别率低。为解决这一问题，提出了一种按文本行重构图像的方法来校正自然扭曲图像。应用形态学膨胀和游程平滑的方法对图像初始化，利用可变长模板搜索样本点，拟合出各文本行中心线，根据中心线估计每行文字的上下位置，利用提出的模型和重构算法进行文本行重构。实验结果表明，校正后图像的OCR识别率能得到显著提高，而且校正速度快，对于1000＊1667像素的图像，校正时间能保证在500毫秒以内。

英文摘要：

When using OCR to recognize document images gotten by the digital camera, the recognition rate is always very low because of the warped content. To solve this problem, a new correction method based on text row reeonfiguration is used to reconfigure each text line. Every middle line of each text row is extracted with morphological expansion and RLSA methods at the beginning. Then using the length alterable model to search text line samples and using these to compute the actual text line equation. At last, each text row＇s top and bottom edges are estimated and the given model is used to reconstruct each text row. The experiment results show that the corrected images＇ OCR rate is significantly improved and correction speed is very fast. For document images of 1000 ＊1667 pixel, the correction time is no longer than 500ms.