基于Mask-RCNN的纸质医药包装钢印字符识别

计算机与现代化 ›› 2022, Vol. 0 ›› Issue (02): 108-113.

基于Mask-RCNN的纸质医药包装钢印字符识别

(1.长沙理工大学物理与电子科学学院，湖南长沙410114；2.纳威尔智能科技有限公司，湖南长沙410007)

出版日期:2022-03-31 发布日期:2022-03-31
作者简介:吴彪(1997—),男,湖南衡阳人,硕士研究生,研究方向:图像处理,计算机视觉,深度学习,E-mail: 1351106367@qq.com；通信作者：周庆华(1977—),男,湖南长沙人,教授,博士,研究方向:人工智能及其应用,电磁波与电磁场理论及应用,E-mail: zhouqinghua@csust.edu.cn；曾小为(1992—),男,湖南长沙人,硕士,研究方向:图像处理,E-mail: 197025991@qq.com。
基金资助:
国家自然科学基金资助项目（42074198）

Stencil Character Recognition of Paper Medicine Packaging Based on Mask-RCNN

(1. School of Physics and Electronic Science, Changsha University of Science and Technology, Changsha 410114, China;
2. Nevil Intelligent Technology Co. Ltd., Changsha 410007, China)

Online:2022-03-31 Published:2022-03-31

摘要/Abstract

摘要： 为实现纸质医药包装钢印字符的实时检测，设计一种基于图像处理和深度学习的钢印字符识别系统。系统首先采用多种图像处理的方法对原始打光下的图像进行预处理，从而自动提取图片中的感兴趣区域，并将其输入训练好的Mask-RCNN网络进行实例分割，得到每张图片中的不同字符的像素位置与其字符数值。实验结果表明，对比传统的字符识别方法，该方法可以很好地解决纸质医药包装钢印字符图片中灰度跳变不明显的问题，准确分割出纸质包装盒图片中的钢印字符并进行标记，其字符的识别准确率达到99%，为生产线上钢印字符的识别和记录提供了新的解决思路，具有较高的实用价值。

关键词: 钢印字符识别, 实例分割, Mask-RCNN, 感兴趣区域, 灰度跳变

Abstract: In order to realize the real-time detection of stencil characters on paper medical packaging, a stencil character recognition system based on image processing and deep learning is designed. The system first uses a variety of image processing methods to preprocess the image under the original lighting, thereby automatically extracting the region of interest in the image, and inputting it into the trained Mask-RCNN network for instance segmentation, then the pixel positions of different characters and their character values in each picture are obtained. The experimental results show that, compared with the traditional character recognition method, this method can solve the problem of insignificant gray-scale jumps in the stencil character pictures of paper medical packaging well, and accurately segment and mark the stencil characters in the picture of the paper packaging box. It has high practical value and its character recognition accuracy rate reaches 99%, which provides a new solution for the recognition and recording of stamped characters on the production line.

Key words: stencil character recognition, instance segmentation, Mask-RCNN, region of interest, gray-scale jump

吴彪, 周庆华, 曾小为. 基于Mask-RCNN的纸质医药包装钢印字符识别[J]. 计算机与现代化, 2022, 0(02): 108-113.

WU Biao, ZHOU Qing-hua, ZENG Xiao-wei. Stencil Character Recognition of Paper Medicine Packaging Based on Mask-RCNN[J]. Computer and Modernization, 2022, 0(02): 108-113.

参考文献

［1］张传果,刘建群. 柱面压印字符识别算法研究和实现［J］. 机床与液压, 2018,46(13):37-41.
［2］ SABET M. 药品生产质量管理规范(GMP)的演变［J］. 流程工业, 2013(22):16-20.
［3］李婧,龚晓峰,王瑞辉. 一种基于多模板匹配的字符识别方法［J］. 微计算机信息, 2009,25(9):258-259.
［4］周国华,商俊燕. 一种抗噪的钢印打码字符识别方法［J］. 计算机与现代化, 2018(12):106-109.
［5］ YE Q X, DOERMANN D. Text detection and recognition in imagery: A survey［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015,37(7):1480-1500.
［6］李建美,路长厚,李学勇. 一种基于图像分层的标牌压印字符分割方法［J］. 光电子．激光, 2008,19(6):818-822.
［7］ HORIUCHI T, TORAICHI K, YAMADA H, et al. Stamped character recognition method using range image［C］// Proceedings of the 2nd International Conference on Document Analysis and Recognition. 1993:782-785.
［8］朱宝伟. 机器视觉中的光源照明设计［J］. 电子科技, 2013,26(3):80-82.
［9］ LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition［J］. Proceedings of the IEEE, 1998,86(11):2278-2324.
［10］KRIZHEVSKY A， SUTSKEVER I， HINTON G. ImageNet classification with deep convolutional neural networks［J］. Communications of the ACM, 2017,60(6):84-90.
［11］GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation［C］// Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition. 2014:580-587.
［12］HE K M, GKIOXARI G, DOLLR P, et al. Mask R-CNN［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020,42(2):386-397.
［13］音松,陈雪云,贝学宇. 改进Mask RCNN算法及其在行人实例分割中的应用［J/OL］. 计算机工程:1-10［2021-04-16］. https://doi.org/10.19678/j.issn.1000-3428.0058058.
［14］REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017,39(6):1137-1149.
［15］SHELHAMER E， LONG J, DARRELL T. Fully convolutional networks for semantic segmentation［J］. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017,39(4):640-651.
［16］GANIN Y, LEMPITSKY V. N4-Fields: Neural network nearest neighbor fields for image transforms［C］// Asian Conference on Computer Vision. 2014:536-551.
［17］侯向宁,刘华春. 基于颜色分割与Sobel算子相结合的车牌定位［J］. 计算机技术与发展, 2018,28(8):156-159.
［18］朱鑫,漆泰岳,王睿,等. 一种改进的用于裂缝图像分割的Otsu方法［J］. 地下空间与工程学报, 2017,13(z1):80-84.
［19］李东兴,高倩倩,张起,等. 融合数学形态学滤波技术的边缘检测算法［J］. 山东理工大学学报(自然科学版), 2018,32(6):1-5.
［20］葛海江,方江江,张翔. 车牌识别系统中车牌定位及倾斜矫正算法研究［J］. 杭州电子科技大学学报, 2007,27(2):49-52.
［21］苗姣姣,徐望明. HSV空间和形态学处理相结合的车牌定位方法［J］. 电视技术, 2015,39(5):146-149.
［22］李武周,余锋,王冰,等. 基于形态学滤波的红外图像背景补偿［J］. 红外技术, 2016,38(4):333-336.

[1]	黄杉杉1, 吴巍2, 徐雨晴1, 魏婕1. 基于改进Mask R-CNN和LSD的图纸管道检测方法[J]. 计算机与现代化, 2024, 0(10): 42-48.
[2]	杨贞, 单孟姣, 殷志坚, 杨凡, 李翠梅. 细粒度图像分类的通道自适应判别性学习方法[J]. 计算机与现代化, 2022, 0(10): 68-74.
[3]	林钦壮1，钟映春1，罗唯师2. 面神经序列图像感兴趣区域识别[J]. 计算机与现代化, 2017, 0(11): 51-54.
[4]	李熙莹1,2,3，卢林1,2,3，余文进1,2,3. 一种基于连通域和跳变的车牌定位方法[J]. 计算机与现代化, 2014, 0(3): 104-108.
[5]	赵新峰;王恒. 一种基于感兴趣区域元数据的视频转码方案[J]. 计算机与现代化, 2012, 1(9): 115-117,.
[6]	范楷;莫建文;张彤;袁华. 基于限定区域和Hough变换的车道线的检测方法[J]. 计算机与现代化, 2011, 1(4): 37-41.