[Cover] [Contents] [Index] |
Page 52
Figure 1.30 Examples of image intensity profiles covered by a window detected by an RGMAP filter. (a) A homogeneous area; (b) and (c) are heterogeneous and only pixels in subwindow S2 are used for estimating local statistics.
(1.62) |
After a series of derivations, the resulting estimate is as follows (Lopes et al., 1990):
(1.63) |
The term is defined in Equation (1.45).
All of the speckle filters described above rely strongly on a good estimate of local statistics (e.g. σz and μz) from a window. If the window centre is located close to the boundary of an image segment (such as a boundary between agricultural fields), the resulting local statistics are likely to be biased and will thus degrade the filtering result. Nezry et al. (1991) note this point, and propose a refined GMAP, called the RGMAP filter, in which the local statistics extracted from a window do not cross image feature boundaries.
The concept of RGMAP is dependent on the ability to detect whether or not the area covered by a window is homogeneous. For a homogeneous area, the local statistics are derived from the whole window. If the area is detected as heterogeneous, the filter uses only a part of window (which is considered homogeneous and of the same feature type as the central pixel)
[Cover] [Contents] [Index] |