A hybrid color space for skin detection using genetic algorithm heuristic search and principal component analysis...

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Color spaces are crucial for image analysis, particularly in skin and face detection.
• Existing color spaces face challenges with illumination variations, camera differences, and diverse skin tones.
• Optimizing color space selection for robust skin detection remains an open research problem.

Purpose of the Study:

• To propose a novel 3D hybrid color space, termed SKN, for enhanced human skin color representation.
• To optimize skin detection accuracy by addressing variations in illumination and skin tones.
• To evaluate the performance of the proposed SKN color space against established methods.

Main Methods:

• Utilized Genetic Algorithm (GA) heuristic to identify optimal color component combinations for skin detection.
• Employed Principal Component Analysis (PCA) to reduce the dimensionality of the GA-optimized solution.
• Evaluated pixel-wise skin detection using four classifiers: Random Forest, Naïve Bayes, Support Vector Machine, and Multilayer Perceptron.

Main Results:

• The proposed SKN color space demonstrated superior pixel-wise skin detection accuracy compared to existing color spaces.
• Using the Random Forest classifier, SKN achieved an average F-score of 0.953 and a False Positive Rate of 0.0482.
• Random Forest was identified as the most suitable classifier for pixel-wise skin detection tasks in this study.

Conclusions:

• The novel SKN color space offers a significant improvement for skin detection applications.
• The combination of GA and PCA provides an effective approach for optimizing color space representation.
• Random Forest classifier is highly effective for pixel-wise skin detection using the proposed SKN color space.