基于对抗生成网络的中国山水画自动生成模型

Chinese Landscape Painting Automated Generation Model Based on Generative Adversarial Networks

中国山水画是以山川自然景观为主要描写对象的画,它是中国画的重要画科。当前深度学习模型在图像分类、对象识别、图像风格转换和图像生成等领域都取得了巨大的成功。提出一个基于深度对抗生成网络的中国山水画自动生成模型,以网络上公开的中国山水画图像为训练集,设计适当深度的网络和损失函数,通过生成器和判别器的对抗训练,得到图像生成器。通过与真实的山水画进行比较,本模型能够生成具有接近中国山水画风格的图像。

Chinese landscape painting mainly describes the natural landscape of mountains and water. It is an important branch of Chinese paintings. Currently deep learning models achieve significant success in many domains, such as image classification, object recognition, image style transformation and automated image generation. In this research, a Chinese landscape painting automated generation model based on generative adversarial networks(GAN) is proposed. The model is trained with Chinese landscape painting images from Internet. The depth of the network and loss function are properly designed. The generator and discriminator are trained in an adversarial manner and finally a well-trained generator is obtained. Compared with true Chinese landscape paintings,the proposed model can generate images with Chinese landscape painting style.