Segmentation, Filtering and Classification of Sar Images

Segmentation plays a key role in many image processing applications, including SAR image analysis. Given a high-quality segmentation, important tasks such as adaptive noise filtering and classification can be greatly simplified. However, the presence of a strong, multiplicative noise known as speckle makes segmentation of SAR images very difficult. A new, hybrid segmentation technique has permitted us to obtain segmentations of good quality prior to filtering. This has led us to develop a region-based version of the LMMSE speckle filter. The segmentations may further be used for the classification of raw or filtered images. In this paper we give an overview of the entire scheme, which segments, filters and classifies SAR images. Sample results on a multitemporal series of ERS-1 images of an agricultural scene are presented.