摘要
针对自然场景图像中多语言文本检测准确率低的问题,提出一种基于计算机的多语言文本区域快速检测方法。首先,根据自然场景中多语言文本文字排列多方向、文字形态多变化、文字大小不统一的特点,选用EAST快速文本检测模型作为多语言文本区域检测方法的基本模型,并分析了该模型结构;然后,为提高EAST模型检测效果,采用VGG16网络改进模型特征提取层,并对模型输出层进行改进设计;最后,通过融合MSER特征与改进EAST模型,并在ICDAR2015、KAIST、MSRA-TD500数据集上进行检测实验,验证了对提出方法的有效性。结果表明,本研究基于计算机的多语言文本区域快速检测方法可有效、准确检测定位自然场景图像中多语言文本区域,相较于EAST模型和改进EAST模型,本研究融合MSER特征与改进EAST模型的文本区域快速检测方法,综合性能提高了约0.02,具有更好的检测效果。
Aiming at the problem of low accuracy of traditional text region detection methods for multi language text detection in real natural scene images,a fast multi language text region detection method based on computer is proposed.Firstly,according to the characteristics of multilingual text in natural scenes,such as multi-directional text arrangement,multi change of text shape and non-uniform text size,East fast text detection model is selected as the basic model of multilingual text region detection method,and the structure of the model is analyzed;Then,in order to improve the effect of East model detection,vgg16 network is used to improve the model feature extraction layer,and the model output layer is improved;Finally,the effectiveness of the proposed method is verified by fusing traditional features with the improved East model and performing detection experiments on icdar2015,KAIST and msra-td500 datasets.The results show that the computer-based fast detection method of multilingual text region can effectively and accurately detect and locate the multilingual text region in natural scene images.Compared with East model and improved East model,the fast detection method of text region based on mser features and improved East model in this study improves the comprehensive performance by about 0.02 and has better detection effect.
作者
冷莉
邹威
LENG Li;ZOU Wei(Sichuan Vocational and Technical College,Meishan Sichuan 620000,China)
出处
《自动化与仪器仪表》
2021年第12期24-27,共4页
Automation & Instrumentation
基金
四川省高职教改项目:高职教育双证融通实施的问题及策略(Z1A6)。
