Title :
k-means classification filter for speckle removal in radar images
Author :
Chen, Honglei ; Kasilingam, Dayalan
Author_Institution :
Dept. of Electr. & Comput. Eng., Massachusetts Univ., North Dartmouth, MA, USA
fDate :
6/21/1905 12:00:00 AM
Abstract :
A new adaptive speckle removal filter for synthetic aperture radar (SAR) images based on a k-means classifier is presented. This filter is able to identify different regions in an image by classifying the image into classes. Speckle is removed by averaging only within a class. This eliminates the effect of smoothing over edges. The filter is shown to preserve edges better than local statistics filters. Performance studies with simulated images of known speckle distributions show that the k-means filter outperforms most existing adaptive speckle removal filters. Simulated images are used to quantify the performance of the filter for single-look and multi-look images. A threshold parameter is defined for 1-look, 4-look and 10-look images. Optimum filter parameters are identified for different image contrasts and speckle noise levels. The filter is also used with real SAR images. The filter is shown to preserve the contrast between different regions while smoothing out the speckle within a region
Keywords :
adaptive signal processing; geophysical signal processing; geophysical techniques; image classification; radar imaging; remote sensing by radar; speckle; synthetic aperture radar; terrain mapping; SAR; adaptive signal processing; edge preservation; geophysical measurement technique; image classification filter; k-means; k-means classifier; land surface; radar image; radar imaging; radar remote sensing; speckle removal; speckle removal filter; synthetic aperture radar; terrain mapping; Adaptive filters; Electronic mail; Image edge detection; Pixel; Radar imaging; Smoothing methods; Speckle; Statistical distributions; Statistics; Synthetic aperture radar;
Conference_Titel :
Geoscience and Remote Sensing Symposium, 1999. IGARSS '99 Proceedings. IEEE 1999 International
Print_ISBN :
0-7803-5207-6
DOI :
10.1109/IGARSS.1999.774592