|
Activity Number:
|
477
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 7, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statisticians in Defense and National Security
|
| Abstract - #302518 |
|
Title:
|
An Efficient Statistical Technique for Automated Band Detection in Remotely Sensed Imagery
|
|
Author(s):
|
Pranab K. Banerjee*+
|
|
Companies:
|
Space Dynamics Laboratory
|
|
Address:
|
1695 North Research Park Way, North Logan, UT, 84341,
|
|
Keywords:
|
Statistical image processing ; band detection ; remote sensing ; pattern recognition ; sensor calibration
|
|
Abstract:
|
Banding in remotely sensed imagery refers to the appearance of brighter or darker regions of fixed width spanning an entire image horizontally or vertically. These artifacts are most common in imagery from push-broom class sensors which are common in space-borne imaging platforms. If any segment of such a sensor suffers from sensitivity/calibration errors, the corresponding pixels will produce banding, and hence their existence is indicative of sensor errors. Because of the enormity of imagery collected by modern sensors, automated detection of these artifacts is essential for assessing sensor health. This paper presents an efficient statistical algorithm for detecting such bands. It requires no a-priori knowledge of their widths or numbers. It is based on single link agglomerative hierarchical clustering in a feature space based on ternary valued "pixel transition vectors."
|