Communications in Information and Systems

Volume 20 (2020)

Number 4

3D facial landmark detection based on differential cylindrical projection and multi-task learning

Pages: 443 – 459

DOI: https://dx.doi.org/10.4310/CIS.2020.v20.n4.a3

Authors

Takuma Terada (Graduate School of Information Science and Engineering, Ritsumeikan University, Shiga, Japan)

Ryusuke Kimura (Graduate School of Medicine, University of the Ryukyus, Okinawa, Japan)

Yen-Wei Chen (Graduate School of Information Science and Engineering, Ritsumeikan University, Shiga, Japan)

Abstract

Facial landmark detection is a fundamental step for face and expression recognition, and identification of personal attributes, analysis of race, and personal authentication. Numerous methods have been proposed for 2D facial landmark detection. However, 3D landmark detection is still a challenging task. In this paper, we propose a 3D facial landmark detection method based on differential cylindrical projection and multi-task learning. We first transform the 3D image to a 2D gray-scale image using cylindrical projection. We further enhance edges and facial parts (i.e. eyes, nose, mouth), which are useful for landmark detection, by using differentiation of the transformed 2D gray-scale image. Then we applied a convolutional neural network to detect the landmarks in the transformed 2D gray-scale image (differential cylindrical projection). Finally, we transformed the detected landmarks back to the original 3D image. Furthermore, we propose to use multi-task learning based on multi-labels pertaining to gender and age to improve detection accuracy. The code is available at: $\href{https://github.com/RU-IIPL/landmark_detection}{\small{\texttt{https://github.com/RU-IIPL/landmark_detection}}}$.

Full Text (PDF format)

Received 30 July 2020

Published 2 December 2020