基于数字图像处理的电能表图像识别技术研究与实现

【作者】 杨娟；

【导师】 陆锦辉；

【作者基本信息】 南京理工大学 ， 通信与信息系统， 2012， 硕士

【摘要】 伴随着科学技术的发展以及国民经济的增长,数字图像处理和模式识别,已经成为科学研究、社会生产中不可缺少的工具,电能表自动识别系统,就是其在生活中的应用。为实现电能表读数和条码的自动识别,设计了一个基于数字图像处理的自动识别系统,主要包括：图像的预处理、目标区域的定位与二值化、图像识别三个部分。在电能表图像预处理阶段,为减小电能表外边框对图像识别的影响,采用了一种基于电能表图像灰度分布去除电表外边框的方法；接着采用中值滤波进行图像平滑,以减小噪声对边缘提取的干扰,然后采用Sobel边缘检测算子检测图像的水平边缘,再进行基于Hough变换的倾斜校正,这种方法具有较高的检测精度和速度。根据条形码黑白条间隔排列的特点,采用扫描线法寻找满足条码跳变特征的区域,实现条码的定位；而对于电能表读数区域,采用边缘检测与投影法相结合的定位方法。为有效地将目标与背景分离开来,在图像二值分割时,结合读数区域的特点,在局部极值二值化(LEVBB)法的基础上,提出了一种改进的自适应阈值的分割算法,对于条码,采用了一种将局部阈值与边缘特性相结合的二值化方法,为下一步图像识别奠定了基础。在图像识别阶段,根据数字字符的先验知识,采用一种改进的投影算法对字符进行分割,然后采用结构特征法进行单个字符的识别。接着分析了数字电表中Code128条码的编码规则,并介绍了条码的识别方法。实验证明,本文的方法取得较好的识别效果,具有一定的可行性。更多还原

【Abstract】 With the development of science and technology and national economy, digital image processing and pattern recognition has become an indispensable instrument in scientific research and production. And the electric meter automatic recognition system is one of its applications in life. In order to get the imformation of the electric meter automaticly, a recognition system, based on image processing technique and image recognition technique, is desided, which consists of 3 steps:image preprocessing; locating and binarization of the target area; image recognition.During image preprocessing, in order to reduce the influence of the electric meter outer border on recognition, a method based on the distribution of image gray is used. Then Median filtering is applied to reduce interference of noise on edge detection, and Sobel edge detection operator is adopted to detect the level edge of the image. After that a method based on Hough transform is used to rectify them. It obtains high detection accuracy and speed.According to the characteristics of the barcode, a scanning line method is selected to find the region which meets the characteristics of the bar code. And a method combining edge detection and projection is used to achieve the location of the number area. In order to separate the target from background effectively, an improved adaptive threshold selection based on the local extreme binary (LEVBB) method is proposed. Furthermore, a general barcode binarization arithmetic based on the characteristics of the bar code is developed, which laid a foundation for further study of image recognition.In the image recognition, according to a priori knowledge of the numbers, an improved methods of projection is used for single character segmentation, and then configuration characteristic algorithm is proposed to recognize the character. Finally this paper analyses the encoding rule of code 128, and Introduces some bar code recognition methods.Experiments show that, this method achieves satisfactory results and have certain feasibility.更多还原