| Title | Histogram estimation for multiple-detector sensors |
| |
| Author | Strome, W M |
| Source | Proceedings of the 6th Canadian Symposium on Remote Sensing; 1980 p. 603-608 |
| Year | 1980 |
| Alt Series | Earth Sciences Sector, Contribution Series 20040252 |
| Alt Series | RESORS 1023775 |
| Meeting | 6th Canadian Symposium on Remote Sensing; Halifax, NS; CA; May 21-23, 1980 |
| Document | book |
| Lang. | English |
| Media | paper |
| Abstract | Many highly efficient Multispectral image analysis algorithms depend upon examination of the shape of the multidimensional probability density function (p.d.f.) of radiance values over a selected
portion of a scene. The p.d.f. is not known, but is usually approximated by the multidimensional histogram. For instruments with a single detector for each spectral channel, such as the Deadalus airborne Multispectral scanner flown by CCRS, this
approximation is adequate. However, where there are many sensors per channel such as is the case with the Landsat MSS difficulties can arise from the essential radiometric correction process. Because each sensor has its own gain and offset
characteristics, the range of input radiometric values corresponding to a given level of quantization at the output is different for each sensor. When the data are radiometrically corrected, the new quantization levels become populated in an uneven
fashion so that the resulting histogram has a series of gaps, "bumps" and "troughs" which are not present in the actual p.d.f. Those analysis algorithms which seek peaks or valleys in the histogram to identify different classes can produce erroneous
results. To counteract this problem, a brute-force approach is taken - namely to discard lower significant bits in the data until a relatively smooth histogram is obtained. This paper will examine an alternative approach in which a histogram is
generated which more closely approximates that which would be obtained from the actual multidimensional radiance p.d.f. |
| GEOSCAN ID | 217054 |
|
|