Joint Segmentation and Identification Feature Learning for Occlusion Face Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Existing occlusion face recognition algorithms often focus on visible facial parts, limiting their performance.
• Current methods rely on external segmentation models, making them sensitive to mask learning accuracy.
• A need exists for robust face recognition systems that can handle occlusions effectively.

Purpose of the Study:

• To propose a joint segmentation and identification feature learning framework for end-to-end occlusion face recognition.
• To develop a novel approach that is less reliant on external segmentation models.
• To improve the accuracy and robustness of face recognition in the presence of occlusions.

Main Methods:

• Developed an integrated framework for joint segmentation and identification feature learning.
• Introduced an occlusion prediction module, supervised by mask labels, that shares features with the identification network.
• Proposed a channel refinement network to generate multi-channel mask matrices and occlusion-free feature maps, followed by feature purification.

Main Results:

• The proposed framework significantly outperforms state-of-the-art methods in both 1:1 face verification and 1:N face identification.
• Experimental results demonstrate superior performance on both synthetic and real-world occlusion face datasets.
• The approach effectively suppresses occlusion elements in both spatial and channel dimensions.

Conclusions:

• The joint segmentation and identification framework offers a more robust solution for occlusion face recognition.
• The novel channel refinement and feature purification modules contribute to improved occlusion handling.
• The large-scale synthetic datasets facilitate better training and evaluation of occlusion face recognition models.