CAS(ME)³: A Third Generation Facial Spontaneous Micro-Expression Database With Depth Information and High Ecological...

Area of Science:

• Computer Science
• Psychology
• Artificial Intelligence

Background:

• Micro-expression analysis (MEA) is crucial for understanding genuine emotions.
• Current MEA methods are limited by small datasets and database bias.
• Developing comprehensive ME databases is challenging.

Purpose of the Study:

• Introduce CAS(ME)³, a large-scale spontaneous micro-expression database.
• Address limitations of existing MEA datasets, including sample size and bias.
• Facilitate advanced MEA research through multi-modal and unsupervised learning approaches.

Main Methods:

• Collected approximately 80 hours of video data with over 8 million frames.
• Manually labeled 1,109 micro-expressions and 3,490 macro-expressions.
• Incorporated depth information, physiological signals, and voice data using the mock crime paradigm.
• Provided unlabeled videos for unsupervised MEA method development.

Main Results:

• CAS(ME)³ offers a large sample size for effective MEA method validation.
• The database supports multi-modal MEA with unprecedented depth information.
• High ecological validity achieved through the mock crime paradigm.
• Demonstrated the effectiveness of depth information with a novel depth flow algorithm.

Conclusions:

• CAS(ME)³ significantly advances micro-expression analysis research.
• The database overcomes previous limitations in sample size and bias.
• Multi-modal and unsupervised approaches are facilitated for practical MEA.