摘要
文字在生活中随处可见,是人们沟通、互动和传递信息的主要媒介。EAST(Efficient and Accuracy Scene Text)模型是一种高效、准确的场景文本检测算法,可以出色地检测自然场景中的文本。但是该方法仍存在长文本检测精度不高等问题。论文对EAST模型进行了改进,在特征提取阶段用Resnet50深度网络来增加模型的鲁棒性,其次在特征融合阶段加入长短时记忆方法LSTM(Long-Short Term Memory),优化了样本信息。实验结果表明,论文设计的场景文字检测算法能够实时处理自然场景图像,准确定位图像中文字的位置,提高了检测精度。
Text is ubiquitous in our daily life and is the main medium for people to communicate,interact and transmit information with each other.EAST is an efficient and accurate scene text detection algorithm that can excellently detect text in natural scenes.However,there are still some problems in this method,such as the detection accuracy of long text is not high.In this paper,the improved EAST algorithm is used to detect text,and the Resnet50 depth network is used to increase the robustness of the model in the feature extraction stage.Secondly,the long-short term memory(LSTM)is added in the feature fusion stage to optimize the text samples.The experimental results show that the scene text detection algorithm designed in this paper can process the natural scene image in real time,accurately locate the position of the text in the image,and improve the detection accuracy.
作者
李玥
束鑫
常锋
LI Yue;SHU Xin;CHANG Feng(School of Computer,Jiangsu University of Science and Technology,Zhenjiang 212003)
出处
《计算机与数字工程》
2021年第9期1753-1757,共5页
Computer & Digital Engineering
基金
国家自然科学基金项目(编号:61572242)资助。
