Target segmentation in active polarimetric images by use of statistical active contours

We address the problem of target segmentation in active polarimetric images, which can reveal contrasts that do not appear in standard intensity images. However, these images are perturbed by strong specklelike noise. For the purpose of segmentation we thus use statistical active contours, which are known to possess noise robustness properties. The polarimetric imagers we consider acquire two different images of the same scene so as to form a two-channel image (TCI). These two images can be combined to form the orthogonal state contrast image (OSCI), which represents the degree of polarization of the backscattered light if its coherency matrix is diagonal. We characterize the segmentation performance of the statistical active contour procedure on the TCI and on the OSCI. In particular, we show that if the illumination beam is spatially nonuniform, it is more efficient to perform the segmentation on the OSCI, which is independent of the spatial variations of the illumination.