TABLE OF CONTENTS Speckle, as seen in the previous chapter, greatly disrupts radar image readability. Due to this, numerous efforts were made to "filter" these images based on two main philosophies. One of them, historically the first to be adopted, aims to improve image appearance by limiting speckle. Standard filters (mean, median, etc.) were used to process SAR images in this approach with the clear goal of bringing radar imaging visually as close as possible to optical images, so that existing methods originally developed for the latter became applicable to the former. The focus then gradually extended from the mere visual aspect to the mathematical problem of estimating reflectivity, i.e. the physical parameter that is characteristic of these objects. "Filters" thus turned into "estimators" taking the specific nature of radar imaging into account. We have moved from noise removal to a different approach aimed at extracting information. The two are of course related, since the images of estimated reflectivity are ideally free of speckle. Further parameters, e.g., the covariance matrix in polarimetric imaging [1] and coherence in interferometry, can also be estimated. In a reflectivity estimation philosophy, the boundary between filtering and segmentation, or even classification, both of which will be discussed in the following chapters, becomes very thin. Indeed, we may find it useful to pair different approaches and resort to, for example, segmentation or structure detection to improve our reflectivity estimates, as will be shown later in this chapter.
[1]
Radar image filtering is a complex problem, given...
