Generalized likelihood ratio test based algorithms for object recognition in photon-limited images

In this paper the problem of detecting and localizing an object embedded in a background image from photon-limited observations is addressed. A new algorithm based on the generalized likelihood ratio test (GLRT) algorithm is formulated and compared to traditional detectors for images in photon-limited noise. We used Monte-Carlo estimation of the localization-receiver-operating characteristics (LROC) curve to evaluate the performance of the proposed algorithm quantitatively and compare it with existing methods. Our experimental results demonstrate that the proposed GLRT approach significantly outperforms traditional photon-limited detectors.