Highlight Removal of Multi-View Facial Images

Area of Science:

• Computer Vision
• Image Processing
• Computer Graphics

Background:

• Specular highlight removal is a long-standing challenge in computer vision.
• Existing methods often focus on single facial images, limiting their applicability.
• Multi-view approaches are needed for more robust highlight removal.

Purpose of the Study:

• To present a lightweight optimization method for removing specular highlight reflections from multi-view facial images.
• To leverage Lambertian consistency for improved highlight removal.
• To provide a method that does not require face reflectance priors.

Main Methods:

• Utilizing Lambertian consistency where diffuse components are view-invariant and specular components are view-dependent.
• Applying non-negative constraints on light and shading in all directions for physical reliability.
• Estimating illumination chromaticity using orthogonal subspace projection.

Main Results:

• A novel dataset with ground truth for multi-view facial highlight removal was created.
• The proposed method demonstrates robustness and accuracy in quantitative evaluations.
• Comparisons show superior performance against existing specular highlight removal techniques.

Conclusions:

• The developed method effectively removes specular highlights from multi-view facial images.
• The approach offers improvements in applications like 3D face reconstruction.
• This work advances the field of highlight removal with a practical and efficient solution.