대립적 네트워크를 사용한 사실적 얼굴 미화 보정 시스템

Abstract

Face beauti cation is one of the most familiar applications of image processing in our daily lives. However, it is hard to obtain natural face photo using commercial image editing tools like Photoshop because beauti cation process heavily relies on users' sense of beauty. This thesis proposes an automated way to generate naturally beauti ed face images. We use a convolutional neural network to extract contextual features for beauti cation of faces. To overcome the lack of paired face images for a beauti cation dataset, we adopt adversarial networks, which makes it possible to conduct an unsupervised training. We also uses a guidance image generated from the input using an example-based method to narrow the search space of the network. Results show that our network has ability to generate naturally beauti ed face images.