计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (2): 164-170.DOI: 10.3778/j.issn.1002-8331.1810-0285

• 模式识别与人工智能 • 上一篇    下一篇

基于FCN的多方向自然场景文字检测方法

杨剑锋,王润民,何璇,李秀梅,钱盛友   

  1. 1.湖南师范大学 信息科学与工程学院,长沙 410081
    2.湖南师范大学 物理与电子科学学院,长沙 410081
  • 出版日期:2020-01-15 发布日期:2020-01-14

Multi-Oriented Natural Scene Text Detection Algorithm Based on FCN

YANG Jianfeng, WANG Runmin, HE Xuan, LI Xiumei, QIAN Shengyou   

  1. 1.College of Information Science and Engineering, Hunan Normal University, Changsha 410081, China
    2.College of Physics and Electronic Science, Hunan Normal University, Changsha 410081, China
  • Online:2020-01-15 Published:2020-01-14

摘要: 传统的自然场景文字检测方法所采用的手工设计特征在应对复杂自然场景时缺乏鲁棒性。针对复杂自然场景中的多方向文字检测问题,提出了一种新的基于深度学习文字检测方法,采用全卷积网络(Fully Convolutional Networks,FCN)并融合多尺度文字特征图,结合语义分割的方法分割文字候选区域,利用分割得到的文字候选区域直接获取文字候选检测框并进行扩大补偿处理,对文字候选检测框进行后处理得到最终检测结果。该方法在ICDAR2013、ICDAR2015标准数据集进行了测评,实验结果表明该方法相比一些最新方法取得了更好的性能。

关键词: 自然场景文字检测, 深度学习, 全卷积网络, 语义分割

Abstract: The traditional natural scene text detection method using some handcraft features always lacks robustness in dealing with complex natural scenes. A novel text detection method based on the deep learning is proposed, which aims at sloving the problem of multi-directional text detection in the complex natural scenes. Firstly, this paper uses Fully Convolutional Networks(FCN) and fuses the multi-scale feature map, it combines semantic segmentation to divide text candidate regions. Secondly, the text candidate detection boxes are obtained through the text candidate regions directly and then it enlarges compensation treatment. Finally, the text detection results are obtained after some post-processings. The proposed method has been evaluated on the ICDAR2013 and the ICDAR2015 standard datasets, and the experimental results show that this method has achieved better performance compared with some state of the art methods.

Key words: natural scene text detection, deep learning, Fully Convolutional Networks(FCN), semantic segmentation