2001 年 40 巻 2 号 p. 6-16
A multipixel, Mueller matrix-based, unsupervised classification algorithm of polarimetric SAR images is proposed. This algorithm is based on a decision formed by the majority of pixel properties derived from the Mueller matrix within a moving window during the classification of the pixel of interest (or the center pixel of the window) . The pixels in an image are classified into three simple scattering types, odd and even number of reflections, and diffuse scattering. This algorithm decreases the unclassified pixels and improves the resultant image smoothness. First, the amplitude scattering matrix is used to calculate the corresponding Mueller matrices for each pixel. The pixel properties of all pixels in a window are then calculated and used to make a decision based on the majority of pixel properties. If all pixels or more than N/2 pixels (or more than 50% of pixels, where N is the number of pixels in a window) in a window are classified into one category, then the pixel of interest will be classified into the same category. Another criterion is added for diffuse scattering. If the pixel properties are completely random, the pixel property of interest will still be placed in the diffuse scattering category. This kind of scattering behavior may occur in forested areas. If a moving window contains no significant pixel property belonging to the categories mentioned above, then the pixel property of interest is unclassified. The window is then moved, and this process is repeated.