Aligning the frames of a non stationary imaging Fourier transform spectrometer for spectrum retrieval

This paper presents an algorithm to align the frames obtained with a non-stationary imaging Fourier transform spectrometer (IFTS). These frames contain both relative motion and intensity variation due to the interference pattern of the IFTS called interferogram. Two motion estimation techniques are combined to register the frames with a sub-pixel precision. An approach based on mutual information is first used for pixel-precision registration with respect to a single reference frame. Sub-pixel precision is then achieved using a phase-based technique. Simulation results show that our registration approach can successfully reconstruct the interferograms of the observed scene. Spectra obtained from these interferograms are undistorted and have better spectral resolution than spectra obtained with a registration algorithm using only phase-based motion estimation, or without motion compensation. We also show that scene elements with different electromagnetic spectrum can be more easily identified in a spectral cube computed from registered interferogram frames.