Robust albedo estimation from a facial image with cast shadow under general unknown lighting

Area of Science:

• Computer Vision
• Image Processing
• 3D Face Reconstruction

Background:

• Facial albedo estimation is vital for 3D morphable models, shape recovery, and illumination-invariant face recognition.
• Current methods often fail due to ignoring cast shadows and relying on statistical models.
• Accurate albedo maps are essential for robust facial analysis.

Purpose of the Study:

• To develop an improved facial albedo estimation method.
• To overcome limitations of existing techniques, particularly concerning cast shadows and unknown lighting.
• To enhance the accuracy of albedo estimation for computer vision applications.

Main Methods:

• A novel method combining image intensity and facial depth information is proposed.
• Albedo estimation is formulated as a linear programming problem minimizing intensity error.
• A secondary step minimizes mean square error of albedo, assuming noisy surface normals derived from depth data.

Main Results:

• The proposed method effectively estimates facial albedo even with cast shadows and unknown lighting.
• Experimental results demonstrate superior performance compared to existing albedo estimation techniques.
• The method provides more accurate albedo maps, crucial for downstream computer vision tasks.

Conclusions:

• The developed method offers a simple yet effective approach to facial albedo estimation.
• It significantly improves upon existing methods by incorporating depth information and addressing shadow issues.
• This advancement holds promise for more robust and accurate facial recognition and 3D modeling.