Phase-Only Circular Harmonics Correlation for Rotation-Invariant Image Tracking

[[abstract]]This paper is mainly concerned with the tracking of targets in time-varying sequences of infrared images, focusing on the effort to eliminate the rotational effect. The major approaches towards this aim include: (1) space- domain, (2) frequency-domain method. Actually, rotation- invariant recognition can be easily realized by circular harmonic (CH) sampling on the image plane and matching in the frequency domain using FFT algorithm. Conventional correlation methods are shift invariant. Their shortcoming is that only the target in a known orientation can be recognized, with a tolerance of several degrees only because they use all the CHs of target as reference. CH filtering is shift invariant and matches to only one of the CH components of the target, resulting in rotation invariance.[[fileno]]2030101030057[[department]]電機工程學