Spectral Image Processing And Recovery For Coded Aperture Hyperspectral Imaging System

Traditionally，imaging systems are designed to obtain the direct spatial-spectralinformation of the target scene. Compressed Sensing (CS) theory has developedrapidly in recent years, which indicates that we can recover high quality image fromless direct measurements through selecting rational design of image measurementsystem. Compressed Coded aperture imaging technique is a novel spatial modulatedtechnique based on CS theory. The imaging technique breaks through the limit of theNyquist sampling theorem, samples the target scene at a low frequency，and recovershigh resolution image through super-resolution reconstruction. In this work, wereport a new Compressive sampling Hyperspectral Imaging (CS-HSI) system basedon Digital Micromirror Device(DMD). In the new CS-HSI system, a DMD is used toimplement the CS measurement patterns, which is inserted into the system tomodulate the intensity of optical image. In decoding, by choosing appropriateoptimization iterative restoration algorithm, we can accurately reconstruct theoriginal high-resolution image from a few low-dimensional projection measurement.The paper focuses on the image restoration algorithms, and the main work is asfollows:The principle of compressed sensing and coded aperture are firstly introducedin the paper，then we set up a hardware frame of CS-HSI system based on DMD，andmake a detailed analysis on the mathematical model and optical principal of thesystem. We discuss the method of system calibration.According to the study of the existing restoration algorithm, the fast iterativeshrinkage threshold algorithm (FISTA) is applied to compressed sensing imagereconstruction，and some improvements are made based on the regularizationparameter. The restoration results based on FISTA indicate that both the imagequality and the convergence speed are improved significantly.We propose a Split Augmented Lagrangian Shrinkage threshold Algorithm(SALSA). The method takes advantage of variable segmentation and the alternatingdirection multiplier. It divides a single spatial domain variable into two spatial andfrequency domain variables, which approach the optimal solution alternately byoptimization algorithm. The restoration results indicate that the algorithm has anevident advantage in both reconstruction quality of image and convergence rate.