Abstract
There are many fields of application for SAR imagery that requires the detection of very thin features. These features often comprise many very small point-like and line-like features. To detect small features from noisy SAR imagery, speckle noise should be reduced prior to high level feature extraction, but features should not be wiped out in the process. To select the most appropriate filter for automated detection of very thin features slightly brighter than the background, the Lee filter, the enhanced Lee filter, the Frost filter, and the small-feature preserving despeckling filter (SFP) developed by one of the authors have been quantitatively evaluated using a ship wake feature in a 6 look ERS-1 image. The ability to reduce speckle and the ability to preserve the contrast of small features were evaluated. To measure the contrast of real images, we developed a new method, in which feature masks and the surrounding background masks are created at multiple threshold levels and the ratios and the differences are calculated from the average pixel values of the feature and the surrounding areas. Among the filters examined, the SFP filter was the most suitable for subsequent automated extraction of very thin slightly bright features.
