Zhuang National Costume Images Generation Method Based On Deep Convolutional Generative Adversarial Network

Abstract

The style design and color matching of Zhuang national costumes is a time-consuming, labor-intensive but important task. For this problem, this paper proposes a method for generating images of Zhuang national costumes based on deep convolutional generative adversarial network. Firstly, combining the strong feature extraction capabilities of the convolutional neural network and generative adversarial network to learn the potential distribution of complex data, a deep convolutional generative adversarial network is designed and constructed. Secondly, for the problem that the generative adversarial network is difficult to converge, Introducing an activation function that is robust to noise and parameter initialization methods for noise, using a discriminator and generator training strategy with an iteration ratio of 1:3 helps the network converge to a steady state. The experimental results show that the method can converge to a stable state and effectively generate colorful Zhuang national costume images.