ML and Bayesian impulse restoration based object recognition in photon limited noise

In this paper two new object recognition and localization approaches are proposed in photon-limited noise. These approaches are based on ML and Bayesian impulse restoration (IR). For the ML approach, the expectation-maximization (EM) algorithm followed by non-linear filtering that highlights impulses is used. For the Bayesian approach, a novel prior that captures the impulsive nature of the desired solution is proposed and an extension of the EM algorithm is used to find the solution. The localization receiver operating characteristics (LROC) curve is used to quantify the performance of the proposed algorithms. Numerical experiments of an extensive Monte-Carlo study are presented. These experiments demonstrate that the proposed ML-IR approach is superior to traditional likelihood ratio detection approaches for this problem. Furthermore, they also demonstrate that the proposed Bayesian-IR framework outperforms its ML counterpart.