Robust Face Alignment by Multi-Order High-Precision Hourglass Network.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Heatmap regression (HR) is a standard technique for face alignment, but struggles with significant pose variations, occlusions, and lighting changes.
• Current HR methods suffer from low-resolution heatmaps and fail to leverage high-order information for discriminative feature learning, limiting performance in challenging scenarios.

Purpose of the Study:

• To develop a robust and high-precision face alignment method capable of handling extreme poses and heavy occlusions.
• To introduce heatmap subpixel regression (HSR) and a multi-order cross geometry-aware (MCG) model to overcome limitations of existing HR approaches.

Main Methods:

• Proposed a novel multi-order high-precision hourglass network (MHHN) integrating HSR and MCG models.
• Developed a subpixel detection loss (SDL) and subpixel detection technology (SDT) within the HSR method for enhanced landmark localization.
• Integrated the MCG model to learn discriminative representations using multi-order cross information, improving facial geometric constraints and context.

Main Results:

• The proposed HSR and MCG integrated within the MHHN achieved state-of-the-art performance on challenging benchmark datasets.
• Demonstrated significant improvements in face alignment accuracy under conditions of extreme pose variations and heavy occlusions.
• The novel approach successfully addressed the limitations of traditional heatmap regression methods.

Conclusions:

• The developed multi-order high-precision hourglass network (MHHN) with heatmap subpixel regression (HSR) and multi-order cross geometry-aware (MCG) model offers a robust solution for high-precision face alignment.
• This work represents the first exploration of heatmap subpixel regression for effective face alignment in unconstrained environments.
• The findings suggest a promising direction for future research in robust facial landmark detection.