Joint sparse learning for 3-D facial expression generation.

Area of Science:

• Computer Graphics
• Machine Learning
• 3-D Computer Vision

Background:

• Realistic 3-D facial expression generation is crucial for film and gaming.
• Existing methods face challenges in efficiency and robustness.

Purpose of the Study:

• To introduce Joint Sparse Learning (JSL) for modeling 3-D face expression relationships.
• To enable effective 3-D face synthesis, retargeting, and restoration.

Main Methods:

• JSL learns mapping functions and their inverses between high-dimensional 3-D faces and low-dimensional representations.
• The approach models relationships for different expressions and identities.
• It also learns mappings for restoring incomplete 3-D face data.

Main Results:

• JSL effectively generates diverse 3-D facial expressions through synthesis and retargeting.
• The method demonstrates robustness in restoring 3-D faces with holes.
• Experimental results show JSL outperforms representative methods in quality and efficiency.

Conclusions:

• JSL provides an effective and efficient solution for 3-D facial expression generation and restoration.
• The proposed method advances the state-of-the-art in realistic 3-D face modeling.