Image Reconstruction Performance Analysis by Optical Implementation in Compressive Spectral Imaging

Spectral images are a collection of detailed data with a large amount of information obtained throughout the different wavelengths. In particular, the acquisition of this volume of data requires expensive instrumentation, high acquisition time, and demanding storage. Currently, compressive sensing algorithms have emerged as an alternative to efficiently acquire and reconstruct images. However, when carrying out a real implementation, a significant drop in performance can be perceived, since compressive sensing algorithms are defined for ideal systems. In this work, it is proposed a detailed study of the different characteristics, conditions, and aberrations that define an optical implemented system, varying sensor size, transmittance, distortions, and additive Gaussian noise. The implementation of a compressive spectral imaging system with misalignment that uses random coded aperture shows that the quality of the reconstructed images has better performance when the system uses a bigger sensor size. Specifically in the case of the sensor of
$64\times 64$
has better performance in a system with misalignment. On the other hand, specifically, it was found that the implementation of the transmittance in calibrated systems with low aberrations affects their performance. In addition, the transmittance value does not affect systems with many aberrations. Also, in the real implementation case, the results showed that the transmittance with better performance was
$T= 25\%$
using a
$32\times 32$
and
$64\times 64$
sensors size.