As-projective-as-possible bias correction for illumination estimation algorithms.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Illumination estimation is crucial for camera auto-white-balance (AWB).
• Current AWB systems often use fast statistical algorithms, which are less accurate than learning-based methods.
• Bias-correction functions can improve the accuracy of statistical illumination estimation.

Purpose of the Study:

• To improve the accuracy of statistical illumination estimation algorithms used in camera AWB.
• To introduce a novel bias-correction function based on an as-projective-as-possible (APAP) transform.
• To demonstrate the effectiveness of the APAP transform with a fast lookup implementation.

Main Methods:

• Building upon prior work that formulated bias correction as a projective transform.
• Developing an as-projective-as-possible (APAP) projective transform that locally adapts to the input R, G, B vector.
• Applying the APAP bias correction to well-known statistical illumination estimation methods.
• Implementing a fast lookup method for the APAP transform.

Main Results:

• The proposed APAP bias correction significantly improves the accuracy of statistical illumination estimation methods.
• The APAP transform offers enhanced performance compared to standard projective transforms for bias correction.
• The fast lookup method enables efficient real-time application of the APAP transform on camera hardware.

Conclusions:

• The APAP projective transform is an effective method for enhancing illumination estimation accuracy in AWB.
• This approach provides a practical way to improve the performance of existing fast statistical algorithms.
• The APAP method offers a viable solution for achieving more accurate auto-white-balance in digital cameras.