At present，supervised learning through convolutional neural networks（CNN）has been widely used in the field of computer vision. In contrast，unsupervised learning with CNN has received less attention. In response to this problem，in order to reduce the application gap between CNN in supervised learning and unsupervised learning，a deep convolution generative adversarial network （DCGAN） is introduced. Using the generative adversarial networks，data training is performed on the existing face dataset，and the representation level of the image local feature to the overall scene in the generator and the discriminator is generated against the network，and the new person face data composed of the individual face features is outputted. By training on various image datasets to make up for the deficiency of image data，the purpose of improving the recognition accuracy is realized，which proves the practicability of the method in unsupervised learning.