
Proceedings Paper
Block-based reconstructions for compressive spectral imagingFormat | Member Price | Non-Member Price |
---|---|---|
$17.00 | $21.00 |
Paper Abstract
Coded Aperture Snapshot Spectral Imaging system (CASSI) captures spectral information of a scene using a
reduced amount of focal plane array (FPA) projections. These projections are highly structured and localized
such that each measurement contains information of a small portion of the data cube. Compressed sensing
reconstruction algorithms are then used to recover the underlying 3-dimensional (3D) scene. The computational
burden to recover a hyperspectral scene in CASSI is overwhelming for some applications such that reconstructions
can take hours in desktop architectures. This paper presents a new method to reconstruct a hyperspectral signal
from its compressive measurements using several overlapped block reconstructions. This approach exploits the
structure of the CASSI sensing matrix to separately reconstruct overlapped regions of the 3D scene. The resultant
reconstructions are then assembled to obtain the full recovered data cube. Typically, block-processing causes
undesired artifacts in the recovered signal. Vertical and horizontal overlaps between adjacent blocks are then
used to avoid these artifacts and increase the quality of reconstructed images. The reconstruction time and the
quality of the reconstructed images are calculated as a function of the block-size and the amount of overlapped
regions. Simulations show that the quality of the reconstructions is increased up to 6 dB and the reconstruction
time is reduced up to 4 times when using block-based reconstruction instead of full data cube recovery at once.
The proposed method is suitable for multi-processor architectures in which each core recovers one block at a
time.
Paper Details
Date Published: 31 May 2013
PDF: 9 pages
Proc. SPIE 8717, Compressive Sensing II, 87170F (31 May 2013); doi: 10.1117/12.2016203
Published in SPIE Proceedings Vol. 8717:
Compressive Sensing II
Fauzia Ahmad, Editor(s)
PDF: 9 pages
Proc. SPIE 8717, Compressive Sensing II, 87170F (31 May 2013); doi: 10.1117/12.2016203
Show Author Affiliations
Claudia V. Correa, Univ. of Delaware (United States)
Univ. Industrial de Santander (Colombia)
Henry Arguello, Univ. of Delaware (United States)
Univ. Industrial de Santander (Colombia)
Univ. Industrial de Santander (Colombia)
Henry Arguello, Univ. of Delaware (United States)
Univ. Industrial de Santander (Colombia)
Gonzalo R. Arce, Univ. of Delaware (United States)
Published in SPIE Proceedings Vol. 8717:
Compressive Sensing II
Fauzia Ahmad, Editor(s)
© SPIE. Terms of Use
