6.1 Introduction

The product model for SAR clutter, introduced in Chapters 4 and 5, postulated that the observed intensity in single-channel images comprises an underlying RCS on which speckle is imposed. This speckling process defines the conditional probability of observing an intensity I from RCS ?, which characterizes the Forward Problem. This chapter is concerned with the Inverse Problem of deriving the RCS, given the observed intensity in the image. Even though speckle properties are well understood, the ambiguity introduced by speckle means that there is no unique inversion from I to ?. However, by interpreting data in terms of a series of increasingly sophisticated models, we can produce image reconstructions of continually improving quality.

This chapter is concerned with the development of filters to provide a reconstruction of RCS. Initially, in Section 6.2, we examine the implications of the speckle model for reconstructed image quality. In Section 6.3 Bayesian inference is introduced, which provides the basis of most of the filters described. Section 6.4 exploits the speckle model alone, with no information about RCS. Once this model has been exhausted, we introduce and exploit a series of further models, commencing with a data model to describe single-point RCS statistics (Section 6.5), followed by world models describing scene properties, namely, neighborhood correlation (Section 6.6) and image structure (Section 6.7). Each improves the quality of the despeckled image. Most of these algorithms are adaptive linear processes. However, we demonstrate in Section 6.8 that nonlinear...