Li Jing, Mahmoodi Alireza, Joseph Dileepan
Innovation Centre for Engineering, University of Alberta, 9211 116 Street NW, Edmonton, AB T6G 1H9, Canada.
Sensors (Basel). 2015 Oct 16;15(10):26331-52. doi: 10.3390/s151026331.
An important class of complementary metal-oxide-semiconductor (CMOS) image sensors are those where pixel responses are monotonic nonlinear functions of light stimuli. This class includes various logarithmic architectures, which are easily capable of wide dynamic range imaging, at video rates, but which are vulnerable to image quality issues. To minimize fixed pattern noise (FPN) and maximize photometric accuracy, pixel responses must be calibrated and corrected due to mismatch and process variation during fabrication. Unlike literature approaches, which employ circuit-based models of varying complexity, this paper introduces a novel approach based on low-degree polynomials. Although each pixel may have a highly nonlinear response, an approximately-linear FPN calibration is possible by exploiting the monotonic nature of imaging. Moreover, FPN correction requires only arithmetic, and an optimal fixed-point implementation is readily derived, subject to a user-specified number of bits per pixel. Using a monotonic spline, involving cubic polynomials, photometric calibration is also possible without a circuit-based model, and fixed-point photometric correction requires only a look-up table. The approach is experimentally validated with a logarithmic CMOS image sensor and is compared to a leading approach from the literature. The novel approach proves effective and efficient.
互补金属氧化物半导体（CMOS）图像传感器的一个重要类别是那些像素响应为光刺激单调非线性函数的传感器。此类包括各种对数架构，它们能够轻松实现视频速率下的宽动态范围成像，但容易出现图像质量问题。为了最小化固定模式噪声（FPN）并最大化光度精度，由于制造过程中的失配和工艺变化，必须对像素响应进行校准和校正。与采用不同复杂度基于电路模型的文献方法不同，本文介绍了一种基于低阶多项式的新颖方法。尽管每个像素可能具有高度非线性响应，但通过利用成像的单调性，可以进行近似线性的FPN校准。此外，FPN校正仅需要算术运算，并且根据用户指定的每像素位数，可以轻松得出最优的定点实现。使用涉及三次多项式的单调样条，无需基于电路的模型也可以进行光度校准，并且定点光度校正仅需要一个查找表。该方法通过对数CMOS图像传感器进行了实验验证，并与文献中的一种领先方法进行了比较。新方法证明是有效且高效的。
