A model for dark current characterization and simulation

Abstract

The Poisson and Normal probability distributions poorly match the dark current histogram of a typical image sensor. The histogram has only positive values, and is positively skewed (with a long tail). The Normal distribution is symmetric (and possesses negative values), while the Poisson distribution is discrete. Image sensor characterization and simulation would benefit from a different distribution function, which matches the experimental observations better. Dark current fixed pattern noise is caused by discrete randomly-distributed charge generation centers. If these centers shared a common charge-generation rate, and were distributed uniformly, the Poisson distribution would result. The fact that it does not indicates that the generation rates vary, a spatially non-uniform amplification is applied to the centers, or that the spatial distribution of centers is non-uniform. Monte Carlo simulations have been used to examine these hypotheses. The Log-Normal, Gamma and Inverse Gamma distributions have been evaluated as empirical models for characterization and simulation. These models can accurately match the histograms of specific image sensors. They can also be used to synthesize the dark current images required in the development of image processing algorithms. Simulation methods can be used to create synthetic images with more complicated distributions.