Parameter estimation is the process of attributing a parametric description to an object, a physical process or an event based on measurements that are obtained from that object (or process, or event). The measurements are made available by a sensory system. Figure 3.1 gives an overview. Parameter estimation and pattern classification are similar processes because they both aim to describe an object using measurements. However, in parameter estimation the description is in terms of a real-valued scalar or vector, whereas in classification the description is in terms of just one class selected from a finite number of classes.
Example 3.1 Estimation of the backscattering coefficient from SAR images
In earth observation based on airborne SAR (synthetic aperture radar) imaging, the physical parameter of interest is the backscattering coefficient. This parameter provides information about the condition of the surface of the earth, e.g. soil type, moisture content, crop type, growth of the crop.
The mean backscattered energy of a radar signal in a direction is proportional to this backscattering coefficient. In order to reduce so-called speckle noise the given direction is probed a number of times. The results are averaged to yield the final measurement. Figure 3.2 shows a large number of realizations of the true backscattering coefficient and its corresponding measurement.1 In this example, the number of probes per measurement is eight. It can be seen that, even ...