Two-Stage Template Matching Using Quantum-Limited Images

Abstract

Two-stage template matching using quantum-limited images is investigated. A high-light-level input scene is searched using a small window function that is moved to various offsets within the input scene. The cross-correlation between the reference function and the input scene at a given offset is estimated by sampling the input scene using a small number of detected photoevents. The photon-limited correlation signal is used as a similarity criterion for comparing the input with the reference function. In the first stage, a small number of detected photoevents is used to find probable locations for the reference function. In the second stage, likely locations are examined using a sufficient number of detected photoevents to provide a recognition decision within a specified error. Results are presented for a simple inspection problem, and for scene matching in a satellite image.