Preprocessing and Feature Descriptors for Facial Micro-Expression Recognition
- Login to Download
- 1 Credits
Resource Overview
Facial micro-expressions reveal crucial insights into human emotions, even when individuals attempt to conceal their feelings. Historically, limited research has been conducted on detecting and recognizing micro-expressions using computer vision techniques. This implementation processes spontaneous micro-expression databases through preprocessing and Haar feature-based image cropping, followed by feature extraction using Local Binary Patterns on Three Orthogonal Planes (LBP-TOP) and Local Gray-Coding Patterns on Three Orthogonal Planes (LGCP-TOP) descriptors. The system employs Support Vector Machines (SVM) for detection and classification, achieving accuracy comparable to existing state-of-the-art methods.
Detailed Documentation
Facial micro-expressions represent subtle changes in how people experience and convey emotions, often revealing underlying feelings even during attempts to mask them. Previous research utilizing computer vision methods for micro-expression detection and recognition remains limited. This implementation demonstrates a pipeline for detecting and classifying spontaneous micro-expressions from databases. The methodology involves preprocessing stages including Haar feature-based image cropping for region localization. For feature extraction, we implement two robust descriptors: Local Binary Patterns on Three Orthogonal Planes (LBP-TOP) which captures dynamic texture patterns across temporal sequences, and Local Gray-Coding Patterns on Three Orthogonal Planes (LGCP-TOP) that encodes grayscale transitions in spatiotemporal domains. The classification phase utilizes Support Vector Machines (SVM) with optimized kernel functions for robust pattern recognition. Experimental results indicate that our approach achieves classification accuracy competitive with contemporary research benchmarks.
- Login to Download
- 1 Credits