基于OCR技术的化验单识别方法研究

发布时间：2018-04-04 17:51

本文选题：OCR　切入点：化验单　出处：《浙江大学》2017年硕士论文

【摘要】：随着医疗互联网的发展,人类医疗产生的健康数据正迅速增长,健康数据体现在就诊后的化验单数据。我国医疗行业医患之间的数量差异导致化验单的解读存在壁垒,高效且准确地解读化验单、管理个人的健康数据是目前健康医疗行业面临的挑战。论文针对解读化验单、管理个人的健康数据中存在的问题,提出基于OCR技术的化验单识别基本流程方法,包括预处理、模式识别、内容识别、识别结果纠错,具体工作如下。首先论文对化验单图像预处理,主要包括二值化、抗扭斜。研究了全局阈值法、自适应阈值法和OTSU方法三种二值化方法,通过对比实验,分析了不同方法的预处理效果,选取OTSU方法作为化验单图像二值化基本方法。随后通过模式识别技术,采用基于霍夫变换的直线检测方法提取化验单图像特征,根据直线特征分类化验单图像、处理化验单图像。再利用Tesseract开源引擎,训练数据、参数调优、识别化验单内容。论文最后对识别结果采取结合编辑距离和化验单医学词库的纠错方案,通过汉字图像相似度比较,确定最终纠错选项。论文通过化验单识别结果纠错前后的对比实验,验证了该纠错方案的有效性。
[Abstract]:With the development of the medical Internet, the health data generated by human medical is increasing rapidly.The difference in the number of doctors and patients in our medical profession leads to the barriers to the interpretation of the laboratory sheet. It is a challenge for the health medical industry to efficiently and accurately interpret the test sheet and manage the individual health data.Aiming at the problems existing in the interpretation of laboratory sheets and the management of personal health data, this paper puts forward the basic flow method of identification of laboratory sheets based on OCR technology, including preprocessing, pattern recognition, content recognition, and error correction of recognition results. The specific work is as follows.First of all, the paper preprocessing the single image, mainly including binarization, torsion resistance.Three binarization methods, global threshold method, adaptive threshold method and OTSU method, are studied. Through comparative experiments, the preprocessing effects of different methods are analyzed, and OTSU method is selected as the basic method of binarization of single image.Then, through the pattern recognition technology, the method of line detection based on Hough transform is used to extract the feature of the laboratory single image, and the single image is classified according to the line feature, and the single test image is processed.Then use Tesseract open-source engine, training data, parameter tuning, identify the content of the test sheet.At the end of the paper, the method of error correction combined with editing distance and laboratory medical lexicon is adopted, and the final error correction options are determined by comparing the similarity of Chinese character images.The validity of the error correction scheme is verified by the contrast experiment before and after error correction.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】