基于DCGAN的手写汉字图像生成研究

Research on image generation of handwritten Chinese characters based on DCGAN

使用计算机生成图像是当前计算机视觉中图像识别研究常用的一种数据增强方法。设计一种基于深度卷积生成对抗网络(DCGAN)的手写汉字图像生成模型。通过消除传统网络中的全连接层,使用批量归一化和反卷积运算来创建深度卷积和反卷积网络结构,并将它们作为生成对抗网络中的判别模型和生成模型来实现对手写汉字图像的生成。实验表明,本设计具有较好的手写体汉字图像的生成效果。

Computer image generation is a data enhancement method commonly used in image recognition research in computer vision.A new image generation model of handwritten Chinese characters based on deep convolution generative adversarial network(DCGAN)is designed.In this model,by eliminating the full connection layer in the traditional network,batch normalization and deconvolution operation are used to create deep convolutional and deconvolution network structures,and they are used as the discriminant model and generation model in the generative adversarial network to realize the generation of handwritten Chinese character images.The experiment shows that this design has a good effect on the generation of handwritten Chinese image.