Skip to Main Content

Image Registration Using Phase Correlation

Masters Thesis, Boise State University, December 2008

Geng Zheng

Abstract:

A phase correlation based image registration method is presented in this thesis to assist in overcoming the disadvantages of the traditional paper based printer test method. The images used in the tests were generated by printer emulators that generate digital images corresponding to those printed on a physical printer. It provides a similarity measure of two given images by computing the correlation between them. This registration method can identify orientational, translational, and scaling misalignments of two images. It combines the basic phase correlation technique and a modified version of it to achieve high speed and high accuracy. 2D phase correlation provides accurate and consistent performance over a wide range of translational misalignments. While registering images using a 1D phase correlation pair is faster, it is only accurate when the translational misalignment is small. The registration method proposed in this thesis speeds up the registration process, while maintaining high accuracy, by downsampling the images, setting a threshold for 2D phase correlation, and using criteria for deciding whether to use 1D phase correlation. The proposed method has been implemented and tested. With the error tolerance set to 5 pixels, over 97% of the image pairs were registered correctly and the average computation time including scale estimation is 1.7 seconds. If there is no scaling misalignment, the computation time can be within 1 second.