during
cy of
Jirector
o thank
D Soft
rk.
n using
sis and
'extural
ystems,
texture
IEEE
e. Vol.
ication,
, edited
ntation
Pattern
| using
n, Vol.
IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring", Hyderabad, India,2002
A SIMPLE TECHNIQUE FOR THE REMOVAL OF VERTICAL COHERENT NOISE
FROM LANDSAT THEMATIC MAPPER IMAGES
Janet Nichol*, Vijay Vohora?
* Department of Land Surveying and GeoInformatics, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong
Department of Earth Science, University of Hong Kong, Pokfulam Road, Hong Kong
Commission VI, WG VI/A
KEYWORDS: Landsat TM, signal-to-noise ratio, median filter, image enhancement
ABSTRACT:
A method is described for removal of a coherent pattern of vertical coherent noise observed on LANDSAT Thematic Mapper images
over homogeneous water surfaces where the image signal is relatively weak. The noise is observable in all six visible wavebands of
the sensor when linear contrast enhancement is applied to improve the visualisation of water quality parameters. Since the noise is
serious enough to affect image interpretation and quantification of these parameters a simple method has been developed for the
removal of the noise using a series of median filters. The results of the noise removal are presented in the form of quantification of
the Signal-to-Noise ratio of the images before and after application of the removal procedures, and suggest a 3 times increase.
1. INTRODUCTION
Increasing concern about human activities in coastal regions
has not been accompanied by the development of fine spatial
resolution satellite sensing systems calibrated for water.
Systems such as SeaWifs, MODIS and MERIS which are
calibrated with a high Signal-to-Noise Ratio (SNR) over water
are more suited to oceanographic applications due to their
coarse spatial resolution. Therefore more detailed studies of
coastlines and small water bodies still rely on sensors having
relatively fine spatial resolution but a low SNR for water, such
as SPOT and Landsat.
In the case of LANDSATSs 4 and 5 Thematic Mapper, noise is
present in the form of striping with a period of 16 lines, due to
gain and offset differences between the 16-row detector scans.
The noise is particularly visible over homogeneous surfaces
such as clouds and water bodies. The magnitude of this noise
has not been considered serious enough to warrant removal
using histogram matching techniques during final product
generation at the ground receiving station due to the large
amount of processing time required (Poros and Petersen, 1985).
Methods have been proposed for removing this 16" line
banding, comprising radiometric equalisation and histogram
matching, as well as various types of kernel-based filtering.
These methods are not completely successful however, because
the noise comprises both a high frequency component with a
radiance variation of up to 5W.m2.ster. and a low frequency
component with radiance variation of up to 15W.m2.ster and
after conventional noise removal the latter is still apparent as a
slow variation in image content over a period of approximately
80 scan lines.
This paper describes the effects of a type of vertical coherent
noise (other than the 16-line banding) which is also present in
LANDSAT TM images over homogeneous surfaces (Figure 1).
This noise comprises coherent banding in a NNE-SSW
direction, with an irregular phase shift between groups of lines.
The period of the pattern is approximately 10-11 pixels and the
amplitude is 1-2 counts. Although of the same magnitude as the
16-line banding, it is not so readily apparent, due to its lack of
linear continuity across the image. This vertical coherent noise
155
has not previously been described in the literature for the
LANDSATS 4 and 5 TM sensors.
Figure 1. Contrast enhanced LANDSAT TM image extract 9*6
km showing boundary of a sediment plume offshore from the
Batanghari River, Sumatra. The transect line refers to Figure 4
However Murphy et al. (1985), evaluating the LANDSAT TM
product received a the Canadian ground receiving station,
identified a coherent noise pattern with maximum amplitudes of
0.5 DN although the magnitude, period and spatial appearance
of the noise was not described. The authors concluded that
since Fourier transforms would be necessary to remove the
noise, corrections should not be made at the receiving station.
The vertical coherent noise discussed in the present paper is
observable in all six visible wavebands of LANDSAT TM
images, and is strong enough, to affect analysis results. It has
been observed by the authors on many different TM images in
coastal regions (eg. Ireland, Singapore, Malaysia, Indonesia,