提出了一种基于多尺度特征提取的场景文本检测方法,用于预测场景中文本的区域.该方法将不同尺度信息加入到网络的特征提取层中,加强了网络结构对尺度变化的鲁棒性.在ICDAR2015和COCO-Text数据集上进行了测试,结果表明该方法均取得了业内领先的成果.%A multi-scale feature extraction-based scene text detection method was proposed for prediction of the text in the scene.This method has the multi-scale information mixed into network's feature extraction layer so as to enhance the robustness of the network to scale variation.Experiments on ICDAR2015 dataset and CO-CO-Text dataset demonstrate that this approach is a new state-of-the-art.
展开▼