ANTI-MASK: AN AUTOENCODER-BASED DEEP NEURAL NETWORK TO REVEAL HIDDEN KOREAN FACES WITH FACE MASKS
- Authors
- Kim, J.; Park, J.W.
- Issue Date
- Feb-2022
- Publisher
- Little Lion Scientific
- Keywords
- Anti-Mask; Autoencoder; Face mask; Image inpainting; Vggface
- Citation
- Journal of Theoretical and Applied Information Technology, v.100, no.3, pp.886 - 894
- Journal Title
- Journal of Theoretical and Applied Information Technology
- Volume
- 100
- Number
- 3
- Start Page
- 886
- End Page
- 894
- URI
- http://scholarworks.bwise.kr/ssu/handle/2018.sw.ssu/42360
- ISSN
- 1992-8645
- Abstract
- The purpose of this study is to draw the face part covered by the mask using deep learning technology to complete the entire image. In this paper, we introduce a method of predicting the lower canal of the face based on the information on the upper part of the face using the autoencoder structure. For this study, we design our anti-mask model based on transfer learning through VGGace and train, evaluate, and experiment with a dataset of 800 Korean frontal faces. Our anti-mask model trained in this way accurately drew a part of the hidden face. Through the evaluation of the drawn face images, we proved that our anti-mask model can sufficiently depict the lower canal from the upper image of the face. Moreover, in this paper, it was demonstrated that drawing by analogy with a part of the face is more accurate than reconstructing by analogy with the entire face. This study is expected to contribute to the development of various applications. © 2022 Little Lion Scientific
- Files in This Item
-
Go to Link
- Appears in
Collections - College of Information Technology > Global School of Media > 1. Journal Articles
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.