Recently, with the rapid growth of security related technologies, interest in security technology using face recognition is increasing. Applying Deep Learning Model to face security technology, security technology based on reliability that can be used in real environment has emerged. However, this technique uses Supervised Learning, which requires a large number of labeled data sets. Producing these datasets requires a lot of time and money, because people have to label them directly. Therefore, we want to create a new image using GAN (Generative Adversarial Networks) which is composed of non-BIA learning. GAN can generate new data according to the learned probability distribution through hostile learning of each module through two models of discriminator of Convolutional Adversarial pair and Generator. In this paper, we try to create a new face recognition dataset through GAN. Celeb A and VGG face dataset ver.2 and dozens of centuries ago used a Western dataset as a learning dataset. The learned data was used to extract various features such as eyes, nose, mouth, etc., to determine if it could be used in a face recognition dataset. Finally, we compare and analyze the difference of each data through histogram analysis for visually similar images.