纸币图像的倾斜会严重影响其中字符识别的精度,为此,需要在字符识别前对纸币图像进行倾斜校正。采用Sobel算子和旋转投影来进行纸币图像的倾斜校正。首先使用Sobel算子和旋转投影来获取倾斜纸币图像的倾斜角度,然后采用双线性插值对纸币图像进行水平旋转校正和垂直错位偏移校正。同时,综合考虑了实际情况和校正的实时性,并规定水平倾斜角和垂直倾斜角都不大于10°。从实验结果来看此方法校正效果较为明显。
Banknote image slant will seriously affect accuracy of character recognition. So, banknote image slant correction need to be done before character recognition. Sobel operator and rotation projection are used to correct slant banknote image. Firstly Sobel operator and rotation projection are used to get inclination angle of banknote image, and then bilinear interpolation is emplazed to carry out level rotation correction and vertical displacement deviation correction. At the same time, considering the actual situation and the real-time of correction, level angle and vertical angle are not more than 10 are defined. From the experimental results that correction effect is more obvious can be seen.