An algorithm is proposed for registering images related by translation, rotation, and scale based on angular and radial difference functions. In frequency domain, the spatial translation parameters are computed via phase correlation method. The magnitudes of images are represented in log-polar grid, and the angular and radial difference functions are given and applied to measure shifts in both angular and radial dimensions for rotation and scale estimation. Experimental results show that this method achieves the same accurate level as classic fast Fourier transform (FFT) based method with invariance to illumination change and lower computation costs.
© 2008 Chinese Optics LettersPDF Article