SparCLeS: dynamic l₁ sparse classifiers with level sets for robust beard/moustache detection and segmentation.

Area of Science:

• Computer Vision
• Biometrics
• Image Processing

Background:

• Facial hair detection and segmentation are crucial for forensic facial analysis.
• Existing methods face challenges with varying illumination and image quality.

Purpose of the Study:

• To propose a novel, fully automatic system (SparCLeS) for robust beard/moustache detection and segmentation.
• To address challenges in detecting facial hair in diverse and difficult facial images.

Main Methods:

• Utilized the multiscale self-quotient (MSQ) algorithm for preprocessing to handle illumination variations.
• Extracted Histogram of Oriented Gradients (HOG) features.
• Developed a dynamic sparse classifier for facial region classification.
• Employed a level set approach for segmenting facial hair regions, integrating global and local image information.

Main Results:

• The SparCLeS system demonstrated effectiveness in detecting and segmenting facial hair.
• Successful validation was achieved across three diverse databases: NIST MBGC, NIST FERET, and LFW.
• The system shows promise for real-world forensic applications.

Conclusions:

• SparCLeS provides a robust and automatic solution for facial hair detection and segmentation.
• The proposed methodology effectively handles challenging facial images.
• This system contributes to advancing soft biometric attribute extraction for forensic analysis.