This manuscript presents a comprehensive methodology to register images taken from multiple cameras mounted on a UAV and triggered nearly simultaneously. Considering the small distance between the cameras relative to the flight height, the registration between images from two cameras could be reduced to a homography transformation and two standard lens distortion corrections. Classical approaches to solve this problem use algorithms to find a set of matching points between the two images. However when cameras capture non correlated bands (like visible and near infrared on vegetation cover), these algorithms fail. We propose a registration approach based on the spatial frequency analysis through a Fourier-Mellin transformation. Each image is partitioned in a set of small rectangular windows, in which the relationship with the other image can be assimilated to a linear transformation. These linear transformations are successively identified by means of Fourier-Mellin transforms, then the hundreds of window links issued are used for further homography and distortion identification for the whole images. The procedure described above has been successfully applied on two sets of wheat crop images with 1cm or 2 cm spatial resolution. The RMS error was always lower than 0.5 pixels, and the maximum error was less than 0.9 pixel. Thus this procedure brings an efficient solution for multispectral image acquisition with a set of cameras.