.png){kind=icon}
.png){kind=icon}
Return
Implicit face model: Depth super-resolution for 3D face recognition
.png){kind=icon}
PRE
AI
DOI：10.1016/j.patcog.2025.111353.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
While 3D face recognition (FR) holds significant promise for real-world applications, the challenge lies in acquiring high-quality depth images. Face super-resolution focuses on reconstructing high-resolution face images from low-resolution ones, yet its application to 3D FR has not been thoroughly studied. In this paper, we design an implicit face model (IFM), demonstrating the feasibility of implicit neural representation for recovering depth faces in a continuous manner. IFM takes coordinates and latent codes distributed in the coordinates of the low-resolution domain as inputs to predict the depth values at the given coordinates of the high-resolution domain. To achieve pixel-level restoration while simultaneously ensuring identity-related details, we propose an attention-guided L1 loss for prioritized recovery of crucial facial structures. A positional encoding strategy, combining Fourier features and coordinate embedding, is introduced to preserve high- frequency details. Extensive experiments validate the effectiveness of IFM in the single-depth super-resolution task and show remarkable 3D FR performance with these super-resolved faces. Our codes are publicly released at https://github.com/wm-bupt/IFM.
Keywords:
Super-resolution
3D face recognition
Implicit neural representation
Depth map
Journal
IF:
7.6
Papers:
1.3W
Citations:
4.5W