Performance Estimation of a Real-Time Rosette Imager

In this paper, we model a real-time feasible rosette imager, consisting of a rosette scanner, an optical sensor and a deterministic image reconstruction algorithm. We fine-tune the rosette imager through selecting the appropriate sensor field of view and rosette pattern. The sensor field of view is determined through a greedy approach using uniform random sampling. Furthermore, the optimal rosette pattern is selected by determining which pattern best covers the imaging area uniformly. We explore image sparsity, image decimation and Gaussian filtering in a well-known natural data set and dead leaves data set using the PSNR, Peak-Signal-to-Noise Ratio. This exploration helps to establish a connection between PSNR and image sparsity. Furthermore, we compare various rosette imager configurations in a Bayesian framework. We also conclude that the rosette imager does not outperform a focal-plane array of equivalent samples in terms of image quality but can match the performance.