Image Registration In Defence

Summary form only given. Much of the work of an image analyst consists of comparing images with previous images taken by the same sensor, to other images collected by different sensors, or even to maps. The changeover from analogue to digital imaging technologies sweeping through Defence opens up the opportunity of providing an analyst with the tools to carry out these comparisons automatically. Skilled image analysts are in very short supply, so anything that increases their throughput is of great due. Automatic image registration is one of these productivity enhancing tools analysts would love to have: the talk presents examples of registration problems occuring in Defence, and describes some of the current research on these problems at DSTO and the Centre for Sensor Signal and Information Processing (CSSIP). Recent breakthroughs in the design of fast hierarchical algorithms make it likely that automatic registration will be able to deliver what analysts and others need. The talk describes two of these breakthroughs. The first is the new class of fast multipole methods: these are making possible high quality interpolation of very large data sets. The second is the general development of wavelet theory. This can be seen as a dramatic expansion of previous approaches to registration based on hierarchical matching and image pyramids: by providing a large body of theory, a wide choice of wavelets and fast decomposition algorithms, wavelet analysis provides a host of new ways to construct algorithms for registration.