Eyeglass Frame Segmentation for Face Image Processing
Kanta Miura, Takamichi Miyamoto, K. Sakurai, Koichi Ito, Takafumi Aoki · 2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · 2022
Many people commonly wear eyeglasses on their face, masking the area around the eyes. The lens part of the eyeglasses can often be visible through the back area, while the frame part of the eyeglasses completely hides the back area, resulting in degrading the performance of face image processing. By taking the eyeglass frame into account in face image processing, we can not only improve the accuracy of recognition and analysis, but also apply it to automatic quality assessment in standardized photos such as passport photos. In this paper, we propose an eyeglass frame segmentation method using the combination of U-Net and PSPNet. We also propose a novel data augmentation method to increase the number of face images with eyeglasses. Through a set of experiments using CelebAMask-HQ, we demonstrate that the proposed method exhibits the efficient performance in the segmentation of eyeglass frames.