Deep CNN Prior Based Image Reconstruction for Multispectral Imaging

Spectral imaging is a widely used diagnostic technique in various fields such as physics, chemistry, biology, medicine, astronomy, and remote sensing. In this work, we focus on a multi-spectral imaging technique with a diffractive lens, which relies on computational imaging, and we develop a novel image reconstruction method that exploits convolutional neural networks. To reconstruct the spectral images from the measurements of the imaging system, the inverse problem is first formulated as a proper optimization problem with regularization. This optimization problem is then divided to subproblems by using the alternating direction method of multipliers, and the subproblem corresponding to a denoising problem is solved with a learning-based convolutional denoising neural network instead of an analytical method. The obtained results illustrate that the proposed method can achieve promising reconstruction performance.