Improved Coregistration of Multidate and Multisensor Orchard Imagery

Combining multi-date and multi-instrument remote sensing images can only yield good results when the images are coregistered very precisely. We present a practical method to improve coregistration of image pairs. It uses area based matching to build a local misregistration model, which is used to warp images to a reference. The challenge is to achieve good matching when images have vastly different appearance due to the instruments or acquisition conditions. The user can avoid errors by masking disturbing areas. Additionally, we propose a sharpness normalization method to facilitate matching. A transformation to an invariant representation is also introduced in order to reduce the effect of shadows. We demonstrate the effect of the methods to improve matching and improve coregistration.