To address the challenges of overall blurriness and structural inaccuracies in few-shot Chinese font generation,a deformable generative network that integrates channel-prior convolutional attention and depthwise separable convolutions is proposed. The network architecture which builds upon the foundation of deformable generative networks is enhanced with channel-prior convolutional attention and depthwise separable convolutions to capture global features better and ensure accurate structural generation of Chinese font characters. Comparative evaluations against four classic models,DG-Font,DG-Font++,MX-Font,and CF-Font,demonstrate that the proposed method excels in metrics such as SSIM (Structural Similarity Index Measure) and FID (Fréchet Inception Distance). Furthermore,ablation experiments validate the effectiveness of the proposed network model. The experimental results demonstrate that the proposed network model effectively mitigates the issues of overall blurriness and structural errors in few-shot Chinese font generation.
关键词
字体生成/小样本/通道先验卷积注意力/深度可分离卷积
Key words
font generation/few shot/channel-prior convolutional attention/depthwise separable convolution
维普
万方数据