66
REMOVAL OF THEMATIC MAPPER STREAKING AND STRIPING ARTIFACTS
Ellis Freedman, Therese Gaines, and Michael Richman
Martin Marietta Corporation
Management and Data Systems
P.O. Box 8555
Philadelphia, Pennsylvania U.S.A 19101
Telephone: (610) 531-1684
Fax: (610) 962-3698
E-Mail: freedmae@land.vf.ge.com
KEY WORDS: Landsat, Thematic Mapper, streaking, striping
ABSTRACT
Some images produced by the Thematic Mappers on Landsats 4 and 5 have displayed a variety of imaging artifacts, including
streaking and striping. Various techniques have been employed to remove these artifacts, but many of these have limitations
and will occasionally induce artifacts of their own. As part of the planned radiometric calibration and correction process being
designed for Landsat 7, a diffuse reflective surface which will obscure the entire aperture will be used as a uniform calibration
target to equalize the output of all of the detectors in each of the reflective spectral bands. Analyses of this procedure with
Landsat 5 Thematic Mapper imagery have provided results indicating not only that the potential exists for eliminating virtually
all visible streaking and striping in Landsat 7, but that correction factors can be generated which may eliminate much of the
streaking in Landsat 4 and 5 as well. Additionally, virtually aii striping can be eliminated through the proper selection and
application of dark shutter information. This paper provides a description of the assumptions made and analyses performed in
evaluating the proposed equalization and destriping algorithms for Landsat 7, the results of the study, and the correction
factors derived for the Landsat 5 Thematic Mapper.
1. INTRODUCTION
The deployment of Landsats 4 and 5 and their associated Thematic Mappers in the early 1980s provided the remote sensing
community with an excellent source of multispectral imagery. However, like many imaging sensors, occasional artifacts
appeared in the imagery which are characteristic of whisk broom systems. Two of the more prominent of these artifacts were
streaking, a systematic darkening or lightening of the output samples from a single detector relative to its neighbors, and
striping (sometimes called banding), a systematic darkening or lightening of the output samples from all of the detectors during
a scan relative to the adjacent scans. The latter is generally the more prevalent.
Thematic Mapper (TM) striping has been associated with a variety of characteristics and names, most often Scan Correlated
Shift, in which the phenomenon is linked to a response difference between the forward and reverse scans. On an absolute
scale, the intensity of striping is small, generally less than 2 digital numbers (counts) out of a maximum possible range of 256.
Yet the human eye is remarkably sensitive to such low intensity patterns. Additionally, the visual effect is amplified when
narrow dynamic range images (water, plains, deserts, etc.) are contrast-stretched to bring out low level details.
Differenttechniques have been developed in an attempt to remove such artifacts (Fischel, 1984). One of the more common of
these is a form of histogram adjustment. Generally, this involves a shifting of the average calibrated output value froma single
scan to match the average calibrated output value of the entire image, and then scaling the standard deviation of each scan to
match that of the entire image. Very often this does improve the visual appearance of theimage. However, the techniques rely
upon certain assumptions about the scene content which are not always true, such as similar statistics at all scan locations,
wide standard deviations, and unimodal distributions. Violation of any of these assumptions may cause the technique to lose
its efficacy, or even to induce its own artifacts.
The Landsat 7 Enhanced Thematic Mapper Plus (ETM+), currently being designed, will have a device which, it is expected,
among other benefits, will provide the means for eliminating much of the potential streaking. The ETM+ will have a calibration
paddle which, when commanded once every few weeks, will swing into a position in front of the aperture, completely blocking
the field of view. On the surface of the paddle facing the aperture will be a highly diffuse, reflective, uniform panel which will
reflect sunlight into the aperture, providing a known (through modeling) radiance with which the reflective bands of the sensor
can be radiometrically calibrated on an absolute basis. Part of the calibration plan calls for equalizing the detector calibration
curves (i.e. ensuring that, given the same input, the calibrated output of each detector is equal) utilizing the extreme uniformity
ofthe panel. Itis expected that once the detector calibration curves have been equalized, the discrepancies in estimating the
calibration curves which can cause streaking, will be eliminated.
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences”, Zurich, March 22-24 1995
sea MN 272 2 2| um PR em
— mn enn IN 0 —
MA CM £033 “Se £m ea — P9 A
M md