Multiresolution-based reconstruction for compressive spectral video sensing using a spectral multiplexing sensor

Abstract

Spectral multiplexing sensors based on compressive sensing attempt to break the Nyquist barrier to acquire high spectral resolution scenes. Particularly, the colored coded aperture-based compressive spectral imager extended to video, or video C-CASSI, is a spectral multiplexing sensor that allows capturing spectral dynamic scenes by projecting each spectral frame onto a bidimensional detector using a 3D coded aperture. Afterwards, the compressed signal reconstruction is performed iteratively by finding a sparse solution to an undetermined linear system of equations. Even though the acquired signal can be recovered from much fewer observations by an  −  -norm recovery algorithm than using conventional sensors, the reconstruction exhibits diverse challenges originated by the temporal variable or motion. The motion during the reconstruction produces artifacts that damages the entire data. In this work, a multiresolution-based reconstruction method for compressive spectral video sensing is proposed. In this way, it obtains the temporal information from the measurements at a low computational cost. Thereby, the optimization problem to recover the signal is extended by adding temporal information in order to correct the errors originated by the scene motion. Computational experiments performed over four different spectral videos show an improvement up to 4dB in terms of peak-signal to noise ratio (PSNR) in the reconstruction quality using the multiresolution approach applied to the spectral video reconstruction with respect to the traditional inverse problem.