摘要:自然场景文本检测与识别研究对于从场景中获取信息有重要意义，而深度学习技术有助于提高文本检测与识别的能力.主要对基于深度学习的自然场景文本检测与识别方法和其研究进展进行整理分类、分析和总结.首先论述自然场景文本检测与识别的相关研究背景及主要技术研究路线；然后，根据自然场景文本信息处理的不同阶段，进一步介绍文本检测模型、文本识别模型和端到端的文本识别模型，并阐述和分析每类模型方法的基本思路和优缺点；另外，列举了常见公共标准数据集以及性能评估指标和方法，并对不同模型相关实验结果进行了对比分析；最后总结基于深度学习的自然场景文本检测与识别技术面临的挑战和发展趋势.



Abstract:Natural scene text detection and recognition is important for obtaining information from scenes, and it can be improved by the help of deep learning. In this study, the deep learning-based methods of text detection and recognition in natural scenes are classified, analyzed, and summarized. Firstly, the research background of natural scene text detection and recognition and the main technical research routes are discussed. Then, according to different processing phases of natural scene text information processing, the text detection model, text recognition model and end-to-end text recognition model are further introduced, in which the basic ideas, advantages, and disadvantages of each method are also discussed and analyzed. Furthermore, the common standard datasets and performance evaluation indicators and functions are enumerated, and the experimental results of different models are compared and analyzed. Finally, the challenge and development trends of deep learning-based text detection and recognition in natural scenes are summarized.



PDF全文下载地址:

http://jos.org.cn/jos/article/pdf/5988