基于ResNet及特征融合的场景文本检测技术研究
Research on Scene Text Detection Based on ResNet and Feature Fusion
论文对于之前的学者所提出的CTPN网络模型进行深入的研究，提出了一种改进的场景文本检测网络。首先，使用ResNet代替原网络中的VGG16来提取层次更深的图片特征。同时，在此基础上对ResNet进行了多层网络特征融合，使提取到的图片特征更加丰富。最后，通过对原有方法的改进，将CTPN的场景文本检测方向由水平拓展为倾斜。在使用相同数据集进行训练的前提下，改进后的网络在ICDAR2011和ICDAR2013中的F1-score值均高于CTPN。
For a long time, natural scene text detection is a very popular research point. In this paper, we study deeply on CTPN proposed by previous scholars, and a improved scene text detection network is proposed. First, we use ResNet instead of VGG16 to extract deeper image features. Simultaneously, On this basis, we fuse the multi-layer network features of ResNet to make the extracted image features richer. Finally, by improving the original method, we expand the direction of scene text detection in CTPN from horizontal to inclined. On the premise of training with the same data set, the F1-score of the improved network in ICDAR2011 and ICDAR2013 is higher than that of CTPN.
牛少彰、李相相
计算技术、计算机技术
人工智能场景文本检测ResNet多层网络特征融合
artificial intelligencescene text detectionResNetmulti-layer network feature fusion
牛少彰,李相相.基于ResNet及特征融合的场景文本检测技术研究[EB/OL].(2020-03-05)[2025-08-24].http://www.paper.edu.cn/releasepaper/content/202003-44.点此复制
评论