A conventional grayscale image sensor pixel can be modeled as a "light bucket." Photons arrive on the sensor pixel for a fixed exposure time. The light bucket fills up with photo-electrons linearly proportional to the number of incident photons until its full well capacity is reached. The final readout is a Poisson random variable (shot noise) plus some Gaussian perturbations (due to quantization, dark current and readout noise). This model, although abstracted away from the underlying analog pixel electronics, is a quite useful model of signal and noise in a conventional image sensor pixel.
I am trying to build a similar signal and noise model, abstracted away from the pixel's MOSFET microelectronics, for a logarithmic image sensor (similar to Section 2 in this paper by Hasinoff et al. or Figure 1b of the EMVA 1288 standard.)
- Is there any notion of saturation for a log pixel?
- How does shot noise come into play in the logarithmic non-linearity?
- Is there any notion of "exposure time" or "integration time" to average out shot noise in a log pixel?
- Is there an EMVA standard (or some other standard) for log sensors (and more generally, for sensors non-linear photo-response, such as a quanta image sensor)?