|Title||An Evaluation of Noise Reduction Algorithms for Data Corrupted by Multiplicative Noise|
|Author||Hirose, T; Pultz, T J; Langham, E J|
|Source||Canadian Journal of Remote Sensing 15, 1, 1989 p. 2-12|
|Alt Series||Earth Sciences Sector, Contribution Series 20041472|
|Media||paper; on-line; digital|
|Abstract||Many spatial filters exist for reducing the multiplicative noise inherent to coherent systems, such as synthetic aperture radar (SAR) and lasers. In certain applications, the noise inhibits the ability
to extract useful information. This holds particularly true for applications requiring the detection of boundaries between different units (e.g., agricultural fields).|
It is well known that increasing the signal to speckle noise ratio (SSNR)
by spatial filtering can significantly improve the detection of boundaries between medium- to high-contract regions. However, the merit of spatial filtering over low-contrast boundaries has not been quantified.
In this study, four spatial
filters are compared for their ability to preserve edges over low- to medium-contrast boundaries. The filters include: the mean, the median, and two adaptive filters. The test imagery consisted of low-contrast areal extended targets simulated by a
chart containing test bars that had varying contrasts between regions and levels of multiplicative noise. The merit of the filters is based on the percentage of pixels within non-edge regions that have been misclassified as a boundary
Results indicate that all the filters perform well over medium-contrast boundaries, with no significant difference in performance. Over medium-/low-contrast boundaries, the difference in merit among the filters increases. At the
lowest contrast, no significant difference is observed.