We have previously proposed a framework containing a typical security camera use case and have discussed how
well this is handled by linear image sensors with various characteristics. The findings were visualized graphically,
using a simple camera simulator generating images under well-defined conditions. In order to successfully render
low-contrast objects together with large intra-scene variations in illuminance, the sensor requirements must
include a high dynamic range combined with a comparably high signal-to-noise ratio. In this paper we reuse the
framework and extend the discussion by including also sensors with non-linear pixel responses.
The obvious benefit of a non-linear pixel is that it generally can cope with a higher scene dynamic range and
that in most cases the exposure control can be relaxed. Known drawbacks are, for example, that the noise level
can be fairly high. More specifically, the spatial noise levels are high due to variable pixel-to-pixel characteristics
and lack of on-chip corrections, like correlated double sampling.
In this paper we ignore the spatial noise, since some of the related issues have been addressed recently.
Instead we focus on the temporal noise and dynamic resolution issues involved in non-linear imaging on a system
level. Since the requirements are defined by our selected use case, and since we have defined a visual framework
for analysis, it is straightforward to compare our findings with the results for linear image sensors. As in the
previous paper, the image simulations are based on sensor data obtained from our own measurements.
The dynamic range is an important quantity used to describe an image sensor. Wide/High/Extended dynamic range is
often brought forward as an important feature to compare one device to another. The dynamic range of an image sensor
is normally given as a single number, which is often insufficient since a single number will not fully describe the
dynamic capabilities of the sensor.
A camera is ideally based on a sensor that can cope with the dynamic range of the scene. Otherwise it has to sacrifice
some part of the available data. For a security camera the latter may be critical since important objects might be hidden
in the sacrificed part of the scene.
In this paper we compare the dynamic capabilities of some image sensors utilizing a visual tool. The comparison is based
on the use case, common in surveillance, where low contrast objects may appear in any part of a scene that through its
uneven illumination, span a high dynamic range. The investigation is based on real sensor data that has been measured in
our lab and a synthetic test scene is used to mimic the low contrast objects. With this technique it is possible to compare
sensors with different intrinsic dynamic properties as well as some capture techniques used to create an effect of
increased dynamic range.