A Study of Calligraphy Font Generation Based on DANet-GAN

In recent years, the generation of calligraphic fonts has set off an upsurge in the field of deep learning research. But the original calligraphers are less preserved due to various problems such as long time. A large number of calligraphic-style fonts are generated through existing authentic relics. Creating a complete set of calligrapher font libraries is of great value for inheriting Chinese culture. At present, all kinds of deep learning methods and traditional methods need a lot of manpower and material resources. And the generation quality is not high. To facilitate the automatic generation of calligraphy characters, this paper proposes a calligraphy character generation system that adds an attention mechanism to the generative antagonism network. In this paper, the Dual Attention Network(DANet) is inserted between the coding layer and decoding layer of the basic zi2zi network. It can improve the generation efficiency of calligraphy characters and generate a large number of calligraphy fonts. In this paper, ablation experiments and comparison tests were carried out on three calligraphy data sets, namely Wen Zhengming, Ouyang Xun, and Liu Gongquan. The generated font structure is complete, the details are clear, and the basic realistic calligrapher's original. The index is better than other models and algorithms, which verifies its effectiveness and universality.