ABSTRACT
Optical elements and image capturing devices can only approximate the scenes that they are used to capture. There are two basic sources or errors in the image capture process, noise and distortion. The topic of this discussion is deterministic distortion and how it is rectified in order to detect objects in a distorted scene. The term distortion refers to either optical distortion or object orientation (i.e., pose) distortion. One approach to detection of distorted images is distortion-invariant correlation filtering. These sophisticated filter designs have been developed, starting in the early 1980s to this day, to compensate for optical and pose distortions. The breadth of distortion literature is vast so we focus our discussion on a generic application of a distortion-invariant object recognition system, which is representative of the basic components of more sophisticated systems.
