水表数字图像的粘连字符识别方法

Effective Two-time Recognition Algorithm of Meter Digital Image

针对水表数字图像中的粘连字符识别问题,提出一种有效的方法。在识别之前,先对水表图像进行前期处理以获得单个字符图像。该方法利用摄像头的安装位置固定,使得字符图像正常大小也基本固定这个先验知识,将分割得到的单个数字图像与正常大小比较,若超过正常大小则认为与边框粘连,将粘连的边框部分切割后再进行模板匹配,取匹配系数最大者为最佳匹配,从而实现粘连字符识别。又考虑到摄像头在实际安装中存在高度上的差异,利用字符图像大小与摄像头高度基本呈线性关系这一先验知识,在识别之前先求出相应高度下的正常字符大小,然后进行粘连识别,最终实现了粘连字符的高度自适应识别。实验结果表明,该方法简单有效且识别率很高,获得了满意的效果。

Contrary to the touching character recognition problem of meter digital images,an effective method is proposed in this paper.Before recognized,meter images are pre-processed to obtain single-character images.This method makes use of the priori knowledge that camera installed in fixed position,which cause the normal size of character images,and compares the size of single-character images separated with normal size.If the size is bigger than the normal size,regard it as a touching image.Then cut the border part of touching,make template matching and take the largest coefficient as the best match to implement touching character recognition.Considering the difference on the camera height in actual installation and taking advantage of the priori knowledge of the linear relationship between character image size and camera height,get the normal character size under the corresponding height and conduct the recognition of touching images.Finally,a highly adaptive recognition of touching character is implemented.