基于改进CTPN的驾驶证打印信息检测方法  

Detection method of driver’s license printing information based on the improved CTPN

作　　者：高刘雅 孙冬[1] 高清维[1] 卢一相[1] 黄志祥 吴丹 GAO Liuya;SUN Dong;GAO Qingwei;LU Yixiang;HUANG Zhixiang;WU Dan(School of Electrical Engineering and Automation,Anhui University,Hefei 230601,China;Key Laboratary of Ministry of Education of Intelligent Computing&Signal Processing,Anhui University,Hefei 230601,China;Anhui Chaoyuan Information Technology Co Ltd,Hefei 230088,China)

机构地区：[1]安徽大学电气工程与自动化学院,安徽合肥230601 [2]安徽大学计算智能与信号处理教育部重点实验室,安徽合肥230601 [3]安徽超远信息技术有限公司,安徽合肥230088

出　　处：《传感器与微系统》2022年第5期114-117,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61402003);安徽高校自然科学研究项目(KJ2019A0022,KJ2019A0023);赛尔网络下一代互联网技术创新项目(NGII20180624)。

摘　　要：为解决目前驾驶证打印中复杂纹理以及原始文字背景条件的限制,提出一种基于改进连接文本提议网络(CTPN)的文本信息完整性检测方法。通过引入帧间差分的方法,先对图像作背景擦除;然后对帧差后的图像进行CTPN文本信息检测,同时引入非极大值抑制(NMS)算法提高检测框的精度。对比未经过背景擦除的图像文本检测结果,实验表明:该方法在驾驶证文本检测中具有更好的文本信息检测效果。In order to overcome the limitation of complex texture and original text background in driver’s license detection,an improved connectionist text proposal network(CTPN) network structure is proposed.CTPN can play an important role in addressing the issue of text detection,firstly,by introducing the method of Frame Difference,we make differentially calculated to the two frames of images at different time and then tests the text information in the images after difference segmentation,and the NMS algorithm is used to improve the accuracy of the text detection.Compared to the result of image text detection without background erasure,experiment shows that the improved method has good detection ability in driver’s license text.

关 键 词：帧间差分 连接文本提议网络 驾驶证识别 文本检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]