基于语义分割的藏文古籍文档文本区域检测

Text Region Detection of Tibetan Historical Documents Based on Semantic Segmentation

藏文古籍文档是中华民族宝贵的文化遗产,采用数字化处理技术对藏文古籍文档进行保护和传承具有重要的历史意义。针对藏文古籍文档文本区域检测提出了一种基于图像语义分割的方法。利用判别式对抗网络框架下的语义分割网络对藏文古籍文档不同类型的文本区域进行了像素分类;其次,根据像素分类结果提取了各个文本区域的轮廓;最后,将检测到的版面布局信息进行保存。最终实现藏文古籍文档的文本区域检测。通过拉萨版《甘珠尔》(藏文)的扫描页面构建了自定义的藏文古籍文档图像数据集,并在上述数据集上进行了实验。经实验表明,在上述数据集上得到了较好的检测效果,证明了上述方法的有效性。

Tibetan historical documents are the precious cultural heritage of the Chinese nation.The use of digital processing technology to protect and inherit Tibetan historical documents has important historical significance.This article proposes a method based on image semantic segmentation for text region detection of Tibetan historical documents.First,using the semantic segmentation network under the Discriminative Adversarial Network framework,the pixel classification of different types of text regions of Tibetan historical documents was carried out.Secondly,the contours of each text region were extracted based on the pixel classification results.Finally,the detected layout information was saved.And finally,the text area detection of Tibetan historical documents is realized.A custom dataset of images of Tibetan historical documents was constructed from the scanned images of the Lhasa version of " bka’’gyur "(Tibetan),and experiments were conducted on this dataset.Experiments show that a better detection effect is obtained on this data set,which proves the effectiveness of the method.