A comparison of image registration techniques for joining sequential images

This paper addresses a comparison of two different methods for registration of a sequential series of images. Such sequences are common, for example, in image sequences acquired from airborne sensors. We compare the two methods from both a theoretical perspective as well as real-world performance. The same error metric based on pixel comparison is used in comparing the two methods. While both registration methods are based on projective transformations, they differ in the manner in which the homographies are computed. We compare their performances and explain the likely cause of the difference.