基于深度学习的藏文现代印刷物版面检测技术研究

Research on Layout Inspection Technology of Modern Tibetan Prints Based on Deep Learning

针对藏文现代图书版面中的文本行分布不均匀、现代藏文字体差异较大的问题,提出了一种基于Faster R-CNN的版面文本行检测算法.通过在整理标注的数据集上训练,用ResNet-50网络提取出藏文现代图书版面特征信息.为了有效提高模型的泛化能力,在COCO数据集下的网络模型中进行迁移学习.实验结果表明,该方法可对藏文现代印刷物的版面实现文本行的定位,检测准确率为83%,召回率为95%,明显提高了版面检测的精确度.

Aimed at the uneven distribution of text lines in the layout of modern Tibetan books and the large differences in modern Tibetan fonts,a layout text line detection algorithm based on Faster R-CNN is proposed.By training on collated and labeled data set,we use the ResNet-50 network to extract the feature information of the Tibetan modern book layout.In order to effectively improve the generalization ability of the model,transfer learning is performed in the network model under the COCO dataset.The experimental results show that this method can realize text line positioning on the layout of modern Tibetan printed materials,with a detection accuracy rate of 83%and the recall rate of 95%,which significantly improves the accuracy of layout detection.