Facial expression-based lie detection: A survey of micro expression analysis, datasets and challenges

This 2025 survey by Hussein and Ali systematically reviews facial micro-expression analysis methods, datasets, and computational challenges for lie detection applications. The paper examines how involuntary, transient facial micro-expressions can reveal concealed emotions during deception, covering micro-expression spotting, recognition, action unit detection, and available benchmark datasets like CASME II and SMIC.

A comprehensive literature survey examining computational approaches to micro-expression analysis for lie detection, reviewing key datasets (CASME, CASME II, SMIC, SAMM), recognition algorithms, and technical challenges in automated deception detection systems.

Background & Context

Micro-expressions are subtle and unconscious behaviors that are usually not easily distinguishable, yet they hold significant promise for deception detection applications. This 2025 survey by Hussein and Ali addresses a critical gap in the polygraph and deception detection literature by systematically reviewing facial micro-expression analysis methods, datasets, and computational challenges.

Micro-expressions are involuntary and transient facial expressions capable of revealing the genuine emotions that people attempt to hide, providing important information in applications such as lie detection and criminal detection. Unlike traditional polygraph methods that measure physiological responses, micro-expression analysis focuses on fleeting facial movements that may leak concealed emotions during deception.

A micro-expression is a brief facial movement which reveals a genuine emotion that a person tries to conceal, and micro-expressions might be unaware and/or uncontrollable to the actor, thus may provide effective clues for detecting lies. This survey emerges at a time when computer vision and machine learning techniques are increasingly applied to automate deception detection.

Research Design & Methodology

As a comprehensive survey paper, Hussein and Ali review the state of micro-expression analysis for lie detection across multiple dimensions. The paper systematically examines computational approaches, benchmark datasets, and technical challenges in automated micro-expression recognition systems.

The survey covers key micro-expression analysis tasks including:

• Micro-expression spotting — detecting the temporal occurrence of micro-expressions in video
• Micro-expression recognition — classifying detected micro-expressions into emotion categories
• Action unit detection — identifying specific facial muscle movements associated with deception
• Dataset evaluation — reviewing available databases for training and testing deception detection systems

Major micro-expression databases like CASME II feature higher temporal resolution (200 fps) and spatial resolution, with 247 micro-expressions selected from nearly 3000 facial movements with action units and emotions labeled. Representative datasets including SMIC (164 micro-expression segments from 16 subjects), CASME II (247 clips from 26 subjects at 200 fps), and SAMM are specifically designed for detection and recognition of spontaneous micro-expressions.

Results & Key Findings

The survey identifies that automated micro-expression analysis shows promise for lie detection but faces significant technical challenges. Current recognition systems achieve variable accuracy depending on dataset characteristics and methodological approaches.

Key findings from the survey literature include:

• Baseline performance using LBP-TOP features and SVM classifiers achieves 63.41% accuracy for 5-class emotion classification on CASME II
• Support vector machines coupled with OpenFace facial feature extraction achieved AUC = 0.72 for videos without cognitive load and AUC = 0.78 with cognitive load applied during interviews
• Muscles related to lie detection are consistent with those with high frequency occurring in micro-expression, and micro-expressions analysis could improve lie detection performance by focusing on the forehead, eyebrows, and cheeks
• Micro-expressions can express seven universal emotions: disgust, anger, fear, sadness, happiness, surprise and contempt

Artificial intelligence performs better than humans in lie-detection tasks, even when humans have more information to make a decision. The survey emphasizes that micro-expressions have important application value in public safety, and since they change rapidly and are difficult to detect, the development of automatic recognition systems has become a research hotspot.

Discussion & Significance

This survey contributes to the polygraph literature by bridging facial behavior analysis with traditional physiological deception detection methods. The paper highlights fundamental differences between micro-expression-based lie detection and conventional polygraph testing—while polygraphs measure autonomic nervous system responses, micro-expressions capture involuntary facial leakage governed by distinct neural pathways.

Two neural pathways control facial behavior: the pyramidal tract responsible for macro-expressions with voluntary facial actions, and the extrapyramidal tract responsible for spontaneous facial expressions. In high-risk scenarios such as lying, both pathways are activated and engage in conflict, resulting in the fleeting leakage of genuine emotions in the form of micro-expressions. This neuropsychological basis provides theoretical support for micro-expression-based deception detection.

The survey's documentation of available datasets and computational methods provides crucial infrastructure for advancing lie detection research. With the rapid advancement of deep learning techniques and increasing interdisciplinary integration, micro-expression recognition is expected to gain broader applications in emotional understanding and social security, with emphasis on developments between 2018 and 2025 including Transformer-based architecture, graph neural networks, and multimodal fusion strategies.

Limitations & Considerations

Most micro-expression recognition methods remain under laboratory control conditions and lack robustness for actual deployment. Light changes, head posture deflection, occlusion, and background interference in natural scenes can significantly weaken recognition effect, necessitating strengthening of systems for "wild environments".

Additional limitations include:

• Micro-expressions are transient and of low intensity, making their detection and recognition difficult and heavily reliant on expert experiences
• Laboratory-based neutral paradigms used in datasets like SMIC and CASME differ from real-world conditions where micro-expressions typically manifest during interpersonal interactions, and individuals may conceal emotions by displaying opposite expressions rather than maintaining neutrality
• Dataset imbalances and limited sample sizes constrain generalization to diverse populations and scenarios
• Ecological validity concerns regarding laboratory-elicited versus naturally occurring deception-related micro-expressions

Practical Applications

For polygraph professionals and security practitioners, this survey provides a roadmap for integrating facial micro-expression analysis with traditional credibility assessment methods. Micro-expression recognition has potential applications in police services and lie detection, teaching assistance, and clinical diagnosis. The documented datasets and computational approaches enable examiners to explore hybrid assessment systems combining physiological sensors with video-based facial analysis.

For consumers and policymakers, the survey highlights both the potential and limitations of automated deception detection systems. While machine learning techniques can distinguish liars from truth-tellers based on facial micro-expressions, and techniques to increase cognitive load during interviews facilitate deception recognition, significant technical and ethical challenges remain before such systems achieve field readiness. The survey underscores the need for continued research on ecological validity, cross-cultural generalization, and real-world performance before micro-expression systems can reliably supplement or replace traditional polygraph testing.