Tensor based structure estimation in multi-channel images

In this work a two-step algorithm is proposed for detecting structures in multi-channel remote sensing images. The first part is a normal low level structure detection performed on the individual images, which provides information of the strength as well as the orientation of the structure. In the second part tensors are used for representing the structure information. This approach has the advantage, that tensors can be averaged either spatially or by applying several images, and the resulting tensor provides information of the average strength as well as orientation of the structure. The performance of the algorithm is examined using multi-frequency, polarimetric SAR images from the Danish airborne EMISAR.