基于DCGAN网络算法的华人人脸生成

Chinese Face Generation Based on DCGAN Network Algorithm

摘要: 近些年来，随着深度学习算法的进一步发展，卷积神经网络进行监督学习在计算机视觉应用上取得了巨大的突破。然而在无监督学习方面，卷积神经网络的关注度并不高。本文将深度卷积增加到无监督学习的生成对抗网络上，用来生成华人人脸。生成对抗网络，一方面可以减少复杂的计算，另一方面生成的图像质量也比较理想。通过使用一万五千张的华人人脸数据集，本文最终生成了具有多样性的华人人脸照片，并且最终的判别器损失函数在0.695上下浮动，生成器损失函数在0.74上下浮动。

Abstract: In recent years, with the extensive research of deep learning algorithms, convolutional neural networks for supervised learning have made huge breakthroughs in computer vision applications. However, in terms of unsupervised learning, the attention of convolutional neural networks is not high. This paper adds deep convolution to the unsupervised learning generative confrontation network to generate Chinese faces. On the one hand, generating a confrontation network can reduce complicated calculations, and on the other hand, the quality of the generated images is also ideal. By using 15,000 Chinese face data sets, this paper finally generates diverse Chinese face photos. The final discriminator loss function fluctuates around 0.695, and the generator loss function fluctuates around 0.74.