基于卷积神经网络的蒙古文字体风格迁移模型

MONGOLIAN FONT STYLE TRANSFER MODEL BASED ON CONVOLUTIONAL NEURAL NETWORK

目前设计一套特定风格样式的蒙古文字体是一项十分繁琐的工作,需要依靠传统人工方式或者计算机辅助对每个蒙古文字体进行单独设计,既耗时又耗力,故一种能自动生成蒙古文字体风格的模型十分有必要。国内外已有学者开展了对汉字和英文字体风格自动迁移的研究,但蒙古文领域仍处于空白阶段。因此,该文提出将卷积神经网络模型应用于蒙古文字体风格迁移,并给出了基于卷积神经网络的蒙古文字体风格自动迁移模型,实现了相应的算法和软件。在蒙古文字体数据集上进行实验,模型采用均方根优化器自动调节深度学习率,逐渐减少差异值,可直接从蒙古文标题字体生成蒙古文手写体等字体,得到的生成字体样式基本接近真实字体样式,达到字体风格迁移的效果。

Designing a set of Mongolian fonts with a specific style is a very tedious task.It relies on traditional manual methods or computer-aided design for every Mongolian font,which is time-consuming and labor-intensive.Therefore,it is necessary to design a model that can automatically generate the Mongolian font style.So many scholars had carried out research on how to transfer the Chinese characters and English font styles automatically,but the Mongolian field was still in a blank stage.The convolutional neural network model was proposed to be applied into the Mongolian style migration in this paper,and the corresponding model was given.Related algorithm and software were carried out as well.Experiments were conducted based on the Mongolian font dataset.The model used the root mean square optimizer to automatically adjust the deep learning rate and gradually reduce the difference values.The Mongolian handwriting font could be directly generated from the Mongolian title font,and the resulting font style was basically close to the real one,reaching the effect of the font style transfer.