Skin color-based video segmentation under time-varying illumination

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Real-time skin segmentation is crucial for various applications.
• Existing methods struggle with dynamic illumination changes.
• Accurate skin color tracking remains a challenge.

Purpose of the Study:

• To develop a novel, robust real-time skin segmentation algorithm.
• To improve segmentation accuracy under diverse lighting conditions.
• To dynamically adapt skin color models over time.

Main Methods:

• Utilized a second-order Markov model to predict skin-color (HSV) histogram evolution.
• Dynamically updated histograms based on segmentation feedback and model predictions.
• Parameterized histogram evolution by translation, scaling, and rotation in color space.
• Employed Maximum Likelihood Estimation for model parameter evolution.

Main Results:

• Achieved up to 24% increase in segmentation accuracy across 17 of 21 test sequences.
• Demonstrated higher skin-color classification rates compared to static models.
• Maintained comparable background classification rates to static methods.
• Showcased reliable performance despite significant illumination variations.

Conclusions:

• The proposed dynamic Markov model approach significantly enhances real-time skin segmentation accuracy.
• The method offers robust performance in challenging, variable lighting scenarios.
• This dynamic approach provides a more adaptive and accurate solution for skin segmentation in video sequences.