基于注意力机制的自然场景文本检测

Text detection in natural scenes based on attention mechanism

针对自然场景文本检测中没有明确全局特征的重要性,导致文本检测过程中存在文本的误检、漏检问题,提出了基于注意力机制的自然场景文本检测方法。该方法在CTPN网络的基础上,利用ResNet网络及特征融合技术提取更深层次的多层网络文本特征;同时将注意力机制引入改进后的特征提取网络中,通过从所有位置聚集的相同特征来增强原始特征,并获取注意力权重,对全局注意力进行汇集,明确需要关注的特征。其次,针对自然场景下文本定位精度不高的问题,使用GIoU损失代替坐标损失,同时引入Focal Loss损失函数对原有损失函数进行改进。实验表明,该方法在自然场景文本图片检测中获得了83%的召回率、87%的准确率和85%的F值,保证了文本检测过程中文本信息的完整性。

Aiming at the fact that the importance of global features is not clear in the text detection of natural scenes,which leads to the misdetection and missed detection of text in the text detection process,a natural scene text detection method based on attention mechanism is proposed.Based on the CTPN network,this method uses the ResNet network and feature fusion technology to extract deeper multi-layer network text features;at the same time,the attention mechanism is introduced into the improved feature extraction network,which is enhanced by the same features gathered from all positions the original features,and the attention weight is obtained,the global attention is collected,and the features that need attention are clarified.Secondly,for the problem of low text positioning accuracy in natural scenes,GIoU loss is used instead of coordinate loss,and the Focal Loss loss function is introduced to improve the original loss function.Experiments show that this method obtains a recall rate of 83%,a precision rate of 87%and an F value of 85%in the text image detection of natural scenes,which ensures the integrity of the text information in the text detection process.