A new method is presented for the fusion of L- and C-band SAR imagery of a scene of sea ice which uses information (in the sense of entropy) computed in each image. It has been shown elsewhere that information can be computed in an image from the conditional probability of intensity differences between adjacent pixels. This information naturally decomposes into structural and textural information. The correlation of structural and textural information in each image separately varies for each identifiable ice type. It is required to find a way to add the information from each image. The local textural information, which is, in essence, a weighted gradient at a point, is computed in two SAR images of similar polarization but differing radar wavelength. The local information from the two images is combined at every pixel using a suggested rule for the addition of an entropy-like measure. The resulting summation is shown to have the same negative exponential probability distribution found for information also. The structure-texture pairs from each of the probability distributions of the fused information for identifiable ice types characterize the ice state of the combined imagery. It is shown that the resulting joint information categories support an identical segmentation to one based on a published technique which uses variance instead of entropy as information.
展开▼
