抄録
This paper describes a new approach for detecting image rotation angles with a high accuracy. Frequency amplitude distributions of input and reference images are transformed to the polar coordinate expression and then the position shift between the polar spectra is detected. In the shift detection, weights depending on the frequency and the amplitude at each frequency are used to precisely calculate the shift. In experiments, four types of synthetic and real image patterns were used as reference images. The average and standard deviation of absolute errors for the rotation mode where synthetic reference images are synthetically rotated are 0.007 and 0.0048 degrees, respectively. Those for other modes where real images are rotated synthetically or with a mechanical stage are within 0.004 and 0.003 degrees, respectively.
