“Improving registration using active shape models and depth” by Bellmore and Canosa

The problem of registering multiple images taken of the same scene is particularly difficult in domains where different sensors are used to collect the information. Registration techniques are used to combine the independent information into a single image for further processing or for additional insight into the data. In video, sequential images are usually highly correlated, making registration simpler; the exception is when a new scene enters the viewfinder or when extreme camera movement results in large differences between sequential images. In these cases, far less information is available for registration. Using multiple sensors trained on the same scene is helpful; however, the precise relationship between the sensors must be obtained to compare the data spatially as well as temporally. In these cases, landmarks are often useful.

1. Cootes, T., and Taylor, C. 1992. Active shape model — smart snakes. In Proceedings of the British Machine Vision Conference, 266275, 1829–1841.