The residual non uniformity of IR detectors is of major concern in the implementation of innovative IR systems. Several
algorithms were developed during the last decade in order to solve this problem. One of these algorithms, "Scene based
non uniformity correction" (SBNUC), is based on the notion that for a moving thermal imager, close by pixels get over
time similar distributions of scene radiation. Following this assumption, differences between the time collected
histograms of pixels are due to non uniformity and can thus be corrected. However, pixels which are not in the closest
proximity of each other need in general more time for their histograms to match. Moreover, depending on the imager
motion characteristics, there can be additional temporal and spatial limitations. An efficient SBNUC algorithm must take
the exact limitations into consideration.
In this work the SBNUC spatial-temporal relations are investigated using the spatial frequency domain representation.
This representation provides an effective point of view since distances in the image are naturally translated into different
spatial frequencies. We show that a way to implement this correction by a recursive time filter incorporates spatial
frequency dependence into the correction speed, allowing the spatial-temporal relation to be engineered easily into the
correction process. Using several characteristic imager motion models we analyze the effect of the motion on the spatialtemporal
relations and demonstrate how an optimal SBNUC process can be designed, for each motion model.
In system performance analysis, most often Signal to Noise Ratio (SNR) and system resolution (via
MTF) are analyzed separately. In this paper we advocate the use of a joint measure, namely, the Noise
Equivalent Reflectance Difference (NERD) as a function of the Spatial Resolution (SR). We
demonstrate that the NERD vs. SR captures most of the essential properties of the system's
performances and is therefore a useful tool in system evaluation. We demonstrate how various
tradeoffs affect the NERD vs. SR curve in some not so trivial way.
Aiming at night time spaceborne imaging, we compare the expected performances of a low-light-level visible sensor
with a conventional IR sensor. The low-light-level visible sensor, an electron multiplier CCD (EMCCD), is a close to
ideal photon counting device, with possibly negligible dark current noise and negligible readout noise. This fact, along
with the significant improvement of diffraction (about an order of magnitude), suggests an interesting competition
between the two technologies. In essence, this is a tradeoff between noise and optical performances (favoring the visible
channel) and basic target radiance (favoring IR). Other factors such as reliability and cost can also play an important role.
While we consider two different spectral ranges with different imaging content, we are able to conduct a cautious
theoretical comparison based on standard targets in various lighting conditions. We show that for a given set of system
parameters, even when lighting conditions are favorable, i.e. a night with a full moon, the low-light-level visible channel
performances are inferior to those of an IR channel. We also comment on the significance of the system working point
regarding performances under varying condition.
Conference Committee Involvement (1)
Infrared Technology and Applications XLV
14 April 2019 | Baltimore, Maryland, United States