在对互感器铭牌图像进行扫描输入时,铭牌图像或多或少会出现一定程度的倾斜,这种图像的倾斜最终会导致其字符识别准确率下降。针对此问题提出一种基于霍夫变换获取图像倾斜角度,进而通过图像旋转矫正提高光学字符识别(Optical Character Recognition, OCR)准确率的方法：首先对原始图像进行二值化,进而获得铭牌的轮廓,再采用基于霍夫变换的方法获得铭牌中的水平线段,通过计算得到线段的水平倾斜角,利用此倾角对图像进行还原。实验结果表明,该方法能快速地计算图像的倾斜角度,提高了OCR识别准确率且准确率可达95%以上。
英文摘要:
In the process of Transformer nameplate image scanning ,images may slant somehow, which may lead to declination of the accuracy rate of character recognition. A method to improve the accuracy of OCR based on Hough Transform has been proposed : First make image binaryzation of the original image, after that we can obtain outline of the nameplate, then acquire the horizontal lines of the nameplate based on Hough Transform, Horizontal angle is obtained by calculation, through this angle we can restore the image. Images that didn’t use rotate correction have very low accuracy rate, it’s easy to lead to failure results. The experimental results show that it’s very fast to calculate slant angle of images and the accuracy rate of OCR has been improved ,which is above 95% when using this method.