场景图像弯曲文本检测与识别  

Detection and Recognition of Curved Text in Scence Image

作　　者：刘航源 LIU Hangyuan(College of Computer Science and Technology,China University of Petroluem(East China),Qingdao 266580)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,青岛266580

出　　处：《计算机与数字工程》2024年第6期1848-1853,共6页Computer & Digital Engineering

摘　　要：随着深度学习技术的发展,场景文本检测与识别在近几年取得了巨大的进步,场景文本检测与识别的实现方式也越来越具有多样性。然而,场景文本的表示方式仍然以矩形和旋转矩形居多,该类文本表示方式难以精准表示弯曲的文本实例。为了解决弯曲文本的表示问题,论文提出了一种参数化的弯曲文本表示法,并且基于参数化表示法设计了弯曲文本纠正算法,实现了弯曲文本的形态纠正,降低了识别的难度。通过引入该方法,实现了功能完善的场景文本检测与识别系统。实验结果表明,此方法可以有效完成场景文本的检测与识别任务。With the development of deep learning technology,scene text detection and recognition has made tremendous prog-ress in recent years,and the implementation methods of scene text detection and recognition have become more and more diverse.However,scene text is still mostly represented by rectangles or rotated rectangles.It is difficult to accurately represent curved text instances in these types of text representation.In order to solve the problem of curved text representation,this paper proposes a para-metric curved text representation method,and designs a curved text correction algorithm based on the parametric representation method,which realizes the morphological correction of curved text and reduces the difficulty of recognition.By introducing this method,a fully functional scene text detection and recognition system is realized.The experimental results show that this method can effectively detect and recognize scene text.

关 键 词：场景文本检测 文本识别 弯曲文本 形态纠正 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]