A pattern recognition technique for scene matching of dissimilar imagery

The problem of accurately aligning the line of sight (LOS) of an infrared imaging seeker with the LOS of a precision pointing and tracking system (PTS) using scene matching techniques is considered in this paper. Several preprocessing algorithms are proposed to transform the two dissimilar video signals to a compatible format. The dissimilarity in the two images is due to the different spatial and spectral responses of the imaging sensors. Several thresholding techniques are presented for converting the eight bit preprocessed video to two levels for input to a binary correlation algorithm. The emphasis of the work was to determine the best trade-offs in order to obtain a real-time correlation algorithm but yet be accurate and reliable. Real-time in this application means that the match point of the two scenes is computed (or updated) on each field of video or at 1/60th second intervals. Simulation results using a particular edge extraction algorithm with several thresholding techniques are given. A new method for improving the correlator (or scene matcher) performance is given.