移动智能终端证件信息识别系统的开发与实现

发布时间：2018-06-19 00:28

本文选题：证件识别 + 图像处理　；参考：《武汉工程大学》2016年硕士论文

【摘要】：传统的信息录入方式是采用人工方式先填写相关表格中信息,再由内部工作人员按照表格内容把关键信息存入计算机,或者是,到指定地点进行证件的扫描上传。前一种方式虽然不限制信息录入的地点,但每一次信息的录入都需要耗费大量的人力物力资源,并且容易出现错误的输入。后一种,虽然在信息录入的效率和准确率上都有提高,但是使用地点却相对固定。移动智能终端的出现,使随时随地进行证件信息的录入成为可能。移动智能终端上的信息识别系统可以广泛的应用于服务性行业、交通系统、公安系统等需要对证件信息进行查验的部分,无需大量人员即可完成证件信息的采集查验,提高采集查验工作中证件信息识别的效率和准确率,具有广阔的应用前景。如何对不同证件中的文字信息进行良好的提取和识别,是开发证件信息识别系统的关键问题。识别一个证件图像的关键信息,首要任务是对其关键信息进行正确提取。本文针对不同证件,设计了不同的图像预处理方法,以确保证件信息能正确提取。本文采用一种字符笔画宽度逼近的二值化方法,对图像进行二值化,减少图像中背景、污点、反光等造成的影响,有效提升信息的识别率。本文在信息识别方面根据不同字符特点,采用了两种目前较为流行的方法对文字进行识别。针对英文数字,本文采用Tesseract-OCR引擎进行识别。英文数字字符结构简单,类别较少,使用Tesseract引擎的识别率已满足本文系统需要,且生成的字符集体积小,满足移动智能终端的使用要求。针对中文汉字,汉字结构复杂且种类众多,使用Tesseract引擎识别率不高,且生成语言体积较大,本文使用一种基于特征提取和卷积神经网络的汉字识别方法,将传统特征提取方法与神经网络结合,弥补了单独使用神经网络训练的过程中丢失的特征信息,并在其每一层使用Dropout技术,有效预防神经网络在训练过程中的过拟合现象,提高最终模型对于文字的识别性能。该方法提升了文字的识别率,且生成模型较小,文字识别速度较快,便于移植到移动智能终端。本文针对以上需求,开发了一款移动智能终端的证件信息识别系统,目前主要支持识别身份证正反面以及行驶证。该系统分为安卓版本和iOS版本,支持市面上绝大多数手机。该系统能成功识别证件上的英文、数字、中文,英文数字识别率在98.4%左右,身份证号码识别率达到99.2%左右,中文识别率达到98.27%左右,证件整体识别率大约为90%。
[Abstract]:The traditional way of information input is to fill in the information in the relevant forms manually, and then the internal staff store the key information into the computer according to the contents of the form, or to the designated place to scan and upload the documents. Although the former method does not limit the location of information input, it requires a lot of human and material resources for each input, and it is prone to the wrong input. Although the efficiency and accuracy of information entry are improved, the location of the latter is relatively fixed. The appearance of mobile intelligent terminal makes it possible to input document information anytime and anywhere. The information identification system on the mobile intelligent terminal can be widely used in the service industry, transportation system, public security system and other parts that need to check the document information, and can complete the document information collection and inspection without a large number of personnel. It has broad application prospect to improve the efficiency and accuracy of document information recognition in collecting and checking work. How to extract and recognize the text information in different documents is a key problem in the development of document information recognition system. To identify the key information of a document image, the most important task is to extract the key information correctly. In this paper, different image preprocessing methods are designed for different documents to ensure that document information can be extracted correctly. In this paper, a binarization method of approaching the width of character strokes is used to binarize the image to reduce the influence caused by background, stain and reflection in the image, and to improve the recognition rate of the information effectively. In this paper, according to the characteristics of different characters, two popular methods are used to recognize characters in information recognition. In this paper, Tesseract-OCR engine is used to recognize English numbers. English numeric characters have simple structure and few categories. The recognition rate of Tesseract engine has met the needs of the system in this paper. The generated character set is small in size and meets the requirements of mobile intelligent terminal. In view of Chinese characters, the structure of Chinese characters is complex and there are many kinds of Chinese characters, the recognition rate of Tesseract engine is not high, and the volume of generated language is large. In this paper, a Chinese character recognition method based on feature extraction and convolution neural network is used. The traditional feature extraction method is combined with the neural network to make up for the missing feature information in the process of training using the neural network alone, and Dropout technology is used in each layer to effectively prevent the phenomenon of over-fitting in the training process of the neural network. Improve the performance of the final model for text recognition. The method improves the recognition rate of characters, and the generated model is smaller, and the recognition speed of characters is faster, so it is convenient to transplant to mobile intelligent terminal. In order to meet the above requirements, a mobile intelligent terminal identification system is developed in this paper. At present, it mainly supports the identification of the positive and negative sides of the ID card as well as the driving card. The system is divided into Android and iOS versions, supporting the vast majority of mobile phones on the market. The system can successfully identify the English, Chinese, Chinese and English numbers on the documents, the recognition rate of the ID numbers is about 98.4%, the identification rate of the ID numbers is about 99.2%, the recognition rate of the Chinese characters is about 98.27%, and the overall identification rate of the documents is about 90%.
【学位授予单位】：武汉工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】