22-09-2012, 02:28 PM
APPLIED NON-UNIFORMITY CORRECTION ALGORITHM FOR STRIPING NOISE REMOVAL IN HYPERSPECTRAL IMAGES
1APPLIED NON-UNIFORMITY.pdf (Size: 6.22 MB / Downloads: 51)
ABSTRACT:
Traditionally, several methods are employed to mitigate striping noise in hyperspectral images, mostly based on histogram, statistical
analysis, and selective frequency filtering such as wavelets. In this work, we present and adapt a successful scene-based nonuniformity
correction algorithm for real-time striping noise reduction. Results are obtained for a set of simulated and real hyperspectral imagery,
with different levels of striping noise. The striping noise removal performance is measured and compared by several metrics, and
the advantages and drawbacks of each method are also discussed. From the results, we conclude that the presented adaptive NUC
based method, for striping noise removal in hyperspectral imagers, present remarkable and competitive results in destriping ability and
radiometrical accuracy.
INTRODUCTION
Fixed pattern noise (FPN) is a serious disturbance common for all
focal plane array (FPA) based imaging systems. It is caused by
the non uniform response of neighbor detectors within the FPA,
producing a grid like degradation in video imagers, and striping
noise in scanning imaging systems such as in push-broom hyper-
spectral imagers in (L. Gmez-Chova and Moreno, 2008). Al-
though this noise problem can be solved by using calibration pro-
cedures, this is not always the case since calibration sources are
not always available, such as in space applications. In addition,
the detectors may suffer from responsivity drifts due to environ-
mental changes, which create the need of performing periodical
calibrations, specially at infrared wavelengths. Fortunately, the
removal of this undesired FPN has been also successfully tack-
led for infrared thermal video cameras by using Non-Uniformtiy
Correction (NUC) techniques. NUC is a calibration procedure
that relies on the online or batch estimation of the detector in-
trinsic parameters that produce the FPN by using the same scene
information gathered by the detector. As the source of the FPN
is the same, the striping noise present on hyperspectral cameras
may also be removed by using NUC techniques, but they may be
adapted to a different image formation model due to the scanning
nature of the acquisition process.
RESULTS
Simulated Data
For the stripping noise simulation we used a multispectral satel-
lite image with three channels (red, green and blue), using a dif-
ferent strip noise pattern with the same statistics for each channel
(μo = 0 and σo = 10%), additionally, a small temporal noise
component was added to simulate the common electronic/thermal
noise. The selected RGB image is show inf Fig. 1.
CONCLUSIONS
The introduced NN NUC correction method over the striping
noise problem has shown promising preliminary results. Even
though the proposed model considers only the offset correction,
it presents better results than the WFT-filtering method. Future
work will be directed to consider a complete correction model
for including the gain parameter, and performing an exhaustive
analysis of the learning rules for improving the estimated process.