Ankur Garg, Meenakshi Sarkar, Dr. S Manthira Moorthi, Debajyoti Dhar

The usual system design compromise in high resolution satellites is to have a higher
spatial resolution panchromatic (PAN) channel with 3 or 4 lower spatial resolution
multispectral (MX) channels and generate a high spatial & spectral resolution prod-
uct on ground by making use of the fusion techniques. Since the telescopes of such
systems are designed to cater to modulation transfer function (MTF) requirements
of higher resolution PAN channel, the MX channels generally have high MTF value
at their Nyquist frequency and hence the data is aliased. Also, the general design in
such systems is to keep along track separation between MX detectors in the focal
plane which results in these channels seeing the same target on ground one after
the other with time separation. Due to this design choice, multiple factors such as
relative change in satellite attitude, effect occurring by differing look angles and
elevation of the terrain, and misalignment occurring due to slight rotation that can
occur between detectors placements of these channels. contribute to sub-pixel shifts
between MX channels during inorbit imaging. This paper proposes an approach to
increase the resolution of MX channels without using any high resolution auxiliary
channel, by exploiting high MTF present in the MX channels and misregistration
arising due to imaging time separation. A new prior which can be used to adap-
tively preserve edges and reduce noise simultaneously, a method which increases the
robustness in point spread function (PSF) estimation from stellar have also been
proposed. Experiments show that the resolution of MX can be improved from 1.6m
to 1m without using any PAN band for ISRO’s Cartosat-2E i.e. self-superresolution.
Our method is generic and can be applied to data from any satellite. The method
is especially useful for increasing the resolution of satellites which have only multi-
spectral channels without a higher resolution panchromatic channel.