Structure-preserving sparse decomposition for facial expression analysis.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Affective Computing

Background:

• Facial expression recognition is challenging due to the complexity of action unit (AU) combinations.
• Existing methods have limited success in utilizing AUs and their compositional rules for accurate expression analysis.

Purpose of the Study:

• To propose a dictionary-based approach for facial expression analysis by decomposing expressions into action units (AUs).
• To develop a method that incorporates domain knowledge and learns structured dictionaries for improved facial expression recognition.

Main Methods:

• Constructing an AU-dictionary using expert knowledge and applying structure-preserving sparse coding.
• Developing a structure-preserving dictionary learning algorithm to learn dictionaries and segment expressive faces into semantic regions.
• Utilizing the computed sparse code matrix for expression decomposition and recognition.

Main Results:

• The proposed dictionary-based approach demonstrates effectiveness in facial expression analysis.
• Experimental results on public datasets validate the method's capability for expression decomposition and recognition.
• The dictionary learning algorithm successfully learns structured dictionaries and semantic facial regions.

Conclusions:

• The dictionary-based approach offers a robust framework for facial expression analysis using action units.
• The proposed methods, including dictionary learning, enhance the accuracy and applicability of AU-based expression recognition.
• This research contributes to advancing the field of automated facial expression understanding.