A taxonomy of two-dimensional image registration techniques is presented based on the types of variations in the images. To fulfill this purpose, the author presents the necessary terminology and image registration fundamentals in the first two sections, and then describes the registration methods in use (section 3) and their basic characteristics (section 4). The registration methods considered have three types of variations: corrected distortions, uncorrected distortions, and variations of interest. The paper presents correlation and sequential methods in subsection 3.1, Fourier methods in subsection 3.2, methods that use point mapping with and without feedback in subsection 3.3, and methods that exploit the behavior of elastic models in subsection 3.4. Each method is characterized by the complexity of its class of transformations determined by the source of misregistration. Both global and local transformations, variations, and computations are discussed.
The material is rich in tables and illustrative figures, and contains a lot of clear explanations. For this complex subject, the length of the paper is just right. The paper contains some misprints, however. For instance, on page 327, a long phrase appears twice, while in the formula for T on page 353, the sign “=” appears in place of “+.”
This paper is a good survey of image registration methods. It has a large and valuable collection of references and is useful for a person working in this area.