Robust Phase Correlation Methods for Sub-Pixel Feature Matching

Precise feature matching is the key technical issue for image registration, moving target tracking and speed measurement, and assessment of changes in environment for UV operation. Fast, accurate and robust functions for image feature matching and comparison formulate the core of machine vision. The recent development of phase correlation method has shown that the technique is capable of measuring sub-pixel level translational displacement between two highly correlated images. However, accuracy, processing speed and reliability are the contradictory factors and none of these factors can be compromised for military operations. This paper presents a robust phase correlation approach achieving reliable sub-pixel image feature matching at high speed. Using both SVD and QMDPE 2-D fitting in combination with fringe filtering, the novel phase correlation engine developed via SEN001 project can measure the translational displacement and rotation between low correlation images at better than 1/10 pixel accuracy. Our experiments also indicate that a small moving target in a large image frame of stationary background (the case of moving target tracking) can be detected and the movement can be measured accurately.