Using polynomials to simplify fixed pattern noise and photometric correction of logarithmic CMOS image sensors

Area of Science:

• Image sensor technology
• Nonlinear signal processing
• Computational imaging

Background:

• Complementary metal-oxide-semiconductor (CMOS) image sensors with monotonic nonlinear responses, like logarithmic architectures, enable wide dynamic range imaging but suffer from image quality issues.
• Pixel response variations due to fabrication mismatch necessitate calibration to minimize fixed pattern noise (FPN) and ensure photometric accuracy.

Purpose of the Study:

• To introduce a novel, efficient, and computationally simple approach for calibrating logarithmic CMOS image sensors.
• To address limitations of existing circuit-based calibration models by employing a polynomial approximation method.

Main Methods:

• Utilized low-degree polynomials for approximating pixel responses, enabling an approximately-linear fixed pattern noise (FPN) calibration.
• Developed a fixed-point implementation for FPN correction using arithmetic operations.
• Employed monotonic splines, specifically cubic polynomials, for photometric calibration without circuit models.
• Derived a fixed-point photometric correction method using a look-up table.

Main Results:

• The novel polynomial-based approach effectively calibrates logarithmic CMOS image sensors.
• Achieved significant reduction in fixed pattern noise (FPN) and improved photometric accuracy.
• Demonstrated computational efficiency and suitability for fixed-point implementation.

Conclusions:

• The proposed polynomial-based calibration method offers an effective and efficient alternative to existing approaches for logarithmic CMOS image sensors.
• This technique simplifies calibration complexity while maintaining high image quality and accuracy.