Speckle reduction for the reconstruction of synthetic aperture radar imagery.

This thesis is aimed towards a deep analysis and synthesis of Kalman filters for reducing the speckle noise in Synthetic Aperture Radar (SAR) images. The motivation stems from an increased number of applications of radar imagery in civilian related projects. For example, in 1996, SAR images acquired from the Canadian RADARSAT satellite were used for environmental monitoring, oil spill detection off the Japanese coast, and floods in northwest United States. Many other useful applications can be mentioned in agriculture, hydrology, geology, forestry, oceanography, etc. Despite its wide usage in the above domains the radar imagery encounters a major drawback due to the quantity of a large speckle component. The purpose of the filters developed in this thesis is the reduction of this component using a Kalman filter in combination with model identification techniques. A justification for using Kalman filters is given. This justification is based, besides a theoretical analysis of the speckle reduction problem, on a set of experiments and tests which are applied to almost all the known filters used in this problem. Special Kalman filters are developed theoretically and applied on practical test and real life images. Special attention is given to the noise component. In the Kalman filtering technique it is well known that small variations in the values of the noise covariances affect the performance. A special chapter is dedicated to the development of an optimal technique for determining the appropriate covariance matrices. Many results and illustrations shown in chapter 7 demonstrate practically the validity of the approach