Performance assessment of reconstruction algorithms for compressed sensing threat detection applications

The novel Compressed Sensing (CS) imaging technique possesses fundamental differences and in special cases outstanding advantages compared to conventional imaging. The crucial requirement is the validity of the ""sparsity"" assumption. Fulfilling this requirement enables the reconstruction of a high-resolution image/signal from very few low-resolution measurements. An excellent example of sparse signals are solar blind UV imaging applications. The image of a hot exhaust plume in front of the dark background is spatially sparse. Hence, a solar blind threat detection system may be considered using a single-pixel-camera equipped with the Compressed Sensing approach. This simple and cheap setup with a spatial light modulator and a single-pixel-detector measures only the brightness of a modulated target scene. It is necessary to reconstruct the original image from the obtained compressed data. Many reconstruction algorithms have been proposed and implemented for various CS applications. However, the demands for a CS threat warning system are very specific. Therefore, we built up a library of common reconstruction algorithms and studied their performance in terms of different image quality assessment metrics. In our investigations we utilized exemplary test scenes as well as a real dataset.