What is Reference (Fixed) Image

The most important component of an image registration algorithm is the similarity metric used to determine when images are in accurate alignment (Penney, Weese, Little, Desmedt, Hill, & Hawkes, 1998). In Figure 1, the inputs to and output from a basic metric are illustrated. In general, a metric works by examining corresponding pixel values in both fixed and moving images and then formulating a measure of similarity based on the relationship between these intensities. The metric assumes that the relationship changes with variations in the spatial transformation used to map between images and a maximum similarity is achieved when the images are in close alignment (Brown, 1992). Intensity equality which is high when pixels are similar is one such relationship employed as a similarity metric in single-modal registration where images are captured using the same sensor type. Total equality, however, is seldom reached due to noise and image acquisition inconsistencies. Additional robustness is therefore achieved by assessing the ratio of intensities and minimising the variance of such ratios. When images are acquired with different sensor types, as is typically the case in multimodal registration, an extension of the ratio method which maximises the weighted sum of variances can be employed. Alternatively, a relationship estimating the entropy of corresponding intensity pairs can be formulated where entropy, derived from information theory (Shannon, 1948), is the measure of the amount of information contained within a signal. Although many algorithms have been proposed, similarity calculation remains a complex task.