Neural network approach for image chromatic adaptation for skin color detection

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Image chromatic adaptation aims to correct for illumination variations, ensuring accurate scene color representation.
• Human skin color detection and adaptation are challenging due to diverse lighting conditions.

Purpose of the Study:

• To develop and evaluate a Neural Network (NN) based method for image chromatic adaptation specifically for human skin color.
• To enable dynamic adaptation to changing illumination conditions for consistent skin color representation.

Main Methods:

• A Neural Network (NN) was trained on diverse color images of human subjects under various illuminations.
• The NN directly predicts illuminant estimates to adapt skin color, achieving color constancy.
• Performance was compared against established methods like Gray World and White Patch.

Main Results:

• The proposed NN method demonstrated effective skin color stabilization across different lighting scenarios.
• Skin regions were successfully detected in NN-stabilized images using an efficient thresholding method.
• Comparative analysis showed advantages over existing color constancy algorithms for skin tone adaptation.

Conclusions:

• Neural Networks offer a promising approach for robust image chromatic adaptation and human skin color analysis.
• The developed method provides a computationally efficient and accurate solution for real-world applications requiring consistent skin color.
• Further research can explore the application of this NN-based technique in broader image analysis tasks.