基于Faster R-CNN及数据增广的满文文档印章检测

针对传统方法过分依赖颜色等特征,导致对古籍文档复印件检测效果不佳的问题,基于深度学习技术建立了一种新的满文文档图像印章检测方法。通过图像变换和合成技术建立满文古籍文档图像数据增广算法解决训练数据不足的问题,在所构建的增广数据集上建立Faster R-CNN深度学习模型挖掘深层图像特征,实现满文文档图像印章检测方法。对采集的真实满文文档复印件图像进行实验,印章检测精度可以达到99.6%,表明本文的方法可以有效的检测古籍文档复印件图像中的印章,对满文文档的研究有重要意义。

Seal Detection on Manchu Document Based on Faster R-CNN and Data Augmentation

Aiming at the problem that the traditional method has excessive dependence on color and other features, which results in poor detection of ancient books, this paper establishes a new Manchu document image seal detection method based on deep learning technology. Firstly, the image data augmentation algorithm of Manchu ancient books is used to solve the problem of insufficient training data through image transformation and synthesis technology. Then, the Faster R-CNN deep learning model is built on the constructed augmented data set to mine the deep image features, and the final Manchu document image seal detection method is realized. Experiment on the copy of the real Manchu document is carried out, and the accuracy of seal detection can reach to 99.6%. It shows that the method in this paper can effectively detect the seal in the image of the ancient document, which is of great significance for the study of Manchu documents.