A study of multitemporal filtering techniques for SAR images speckle reduction

摘要

Our purpose is to study speckle filtering of multitemporal SAR images. This communication presents the results of the research undertaken in our laboratory. Different methods of quality restoration on amplitude SAR images are studied and different kinds of SAR multitemporal filters are proposed: the first one is based on a weighted average of each different acquisition. This filter is proposed by Stroobants and takes the temporal variation into consideration. It is simply based on a weighted average between the value of each temporal brother of a given pixel. The second describes two versions of the filter proposed by Quegan. In this case, the idea is to linearly combine M images from the same scene to produce M speckle reduced images. A linear intensity multitemporal filter based on the covariance matrix can combine images to achieve better speckle reduction, while preserving spatial resolution of individual channels. The covariance matrix plays an important role in this filter. When data are uncorrelated, the filter takes its simplest form. The last filter studied in this communication is based on the filter proposed by the team of Alparone. The performance of these filters increases with the number of images used in filtering. The filtering effect can be quantified using ENLs measured from visually homogeneous regions in the images, bias and edge detection performance. A qualitative and quantitative comparison of all developed methods is achieved for four ERS-1 coregistered multitemporal SAR images of the Algerian region including urban areas of the town of Algiers. The encouraging results obtained, the improvements with respect to the classic algorithms of speckle filtering and the fundamental limits on filters performance are presented and discussed.