利用多种投票策略的水表读数字符分割与识别

基于安卓平台提出了一种对水表图像读数进行自动识别的新方法。水表读数字符分割和识别算法主要包括三部分:(1)利用霍夫变换和一种投票策略检测出水表的外部轮廓,然后基于模板影像和水表中四个红色指针位置,利用仿射变换技术对水表字符区域进行定位;(2)对图像进行预处理和连通区检测,提出一种基于投票策略的方法进行字符分割;(3)在考虑数字字符可能出现粘连噪声和缺失的情况下,使用改进的Tesseract OCR识别引擎对字符进行识别。最终通过NDK编译平台将实现的C++程序移植到Android环境中。经过对100组数据进行系统功能测试,结果表明,水表智能识别系统取得了一定效果,能对水表读数进行比较准确和快速的读取。

Water-Meter Character Segmentation and Recognition Method by Using Several Voting Strategies

An automatic water-meter character recognition method for the water-meter image obtained by the Android mobile phone was proposed. The proposed water-meter character segmentation and recognition algorithm mainly include three parts: (1) the Hough transform and a voting strategy were used to detect the external outline of the water-meter, and then the affine transformation technique was used to locate the character region based on the template image and the positions of the four red pointers in the water-meter. (2) After doing some image preprocessing, the connected regions in the image were detected, and then the character segmentation based on another voting strategy was completed. (3) In the case of considering the possible adhesion noise and the partial loss of the characters, the improved tesseractOCR recognizer was used to do the character recognition. The completed C++ program of the water-meter character recognition system is ported to the Android environment through the NDK platform. Experiments conducted on 100 groups of test images showed that our water-meter character recognition system can achieve a certain effect and it can automatically read the water-meter characters accurately and quickly.