Log-Polar Optical Coordinate Transformation with Applications for Automatic Pattern Recognition.

摘要

An important task in machine vision and target recognition is the rapid estimation of the orientation and size of an object with respect to a reference. An optical image processing system is described which converts in-plane rotation and size variations into shift properties using log-polar coordinate transforms. The optical system consists of a binary optical element, which performs the log-polar coordinate transform of the input image, a Fourier transform lens, a Michelson interferometer and a CCD camera for collecting the scale/rotation invariant output. The Michelson interferometer produces two identical patterns on the polar axis to prevent wraparound effects of the coordinate transform. Results are presented in terms of capability in estimating the rotation and size of simple images and discriminating between similar in-class and out-of-class images using the output of a binary phase-only correlator. Pattern recognition, Phase-only filters, Image processing. Space-variant sensing.