A Fast Alternating Minimization Algorithm for Coded Aperture Snapshot Spectral Imaging Based on Sparsity and Deep Image Priors

TL;DR

A fast alternating minimization algorithm based on the sparsity and deep image priors of natural images based on the principle of compressive sensing reconstruction is proposed to significantly improve the accuracy of reconstruction.

Abstract

Coded aperture snapshot spectral imaging (CASSI) is a technique used to reconstruct three-dimensional hyperspectral images (HSIs) from one or several two-dimensional projection measurements. However, fewer projection measurements or more spectral channels leads to a severly ill-posed problem, in which case regularization methods have to be applied. In order to significantly improve the accuracy of reconstruction, this paper proposes a fast alternating minimization algorithm based on the sparsity and deep image priors (Fama-SDIP) of natural images. By integrating deep image prior (DIP) into the principle of compressive sensing (CS) reconstruction, the proposed algorithm can achieve state-of-the-art results without any training dataset. Extensive experiments show that Fama-SDIP method significantly outperforms prevailing leading methods on simulation and real HSI datasets.

Figures (7)

• Figure 1: Schematic of the CASSI system.
• Figure 2: The system matrix $\mathbf{H}$. The figure depicts the sensing of three spectral bands and two snapshots.
• Figure 3: Schematic diagram of the neural network structure used for solving $\textbf{Step 2}$. The input is progressively downsampled by factor of 2 at each scale (e.g. $M_4 = M_1 / 8$).
• Figure 4: Spectral data scenes from (a) ICVL and (b) KAIST data sets used in simulations.
• Figure 5: Reconstructed simulation HSIs comparisons of Scene 7 and 9 with 4 out of 31 spectral channels. The reconstructed spectral curves on selected regions are shown for comparing the spectral accuracy of different algorithms. The correlation of the reconstructed spectra is shown in the legends.