THIS SITE

Disparate Scene Registration

A problem of great interest in the field of visual pattern recognition is image registration, where images of the same scene captured under different conditions are aligned with each other. Image registration is important to a wide range of applications, such as environment change analysis, superresolution, biomedical image analysis, and computer-assisted surgery.  Much of recent research in image registration has focused on multimodal image registration, where images captured using different imaging modalities are registered together.

Multimodal image registration is a very challenging problem for many reasons. Images captured using different imaging modalities are represented using different intensity mappings, making it difficult to compare images in a direct manner based on intensity values. This situation is further complicated by the presence of signal non-homogeneities, noise, and geometric distortions. Furthermore, there are many cases where it is necessary to register multiple images from different imaging modalities to obtain a more complete picture of the scene (e.g., T1/T2/CT registration, LANDSAT 4/5/7 inter-band registration). Hence, it is important to maintain internal consistency between all the images being registered, which is a difficult task to accomplish in the case of multimodal registration given the different information being captured. Therefore, an automatic method that can address all of the above issues is highly desired.

In the VIP lab, we investigate simultaneous, invariant strategies for tackling the problem of registering multiple disparate images acquired using different imaging modalities. In particular, we explore the concept of joint complex phase order distributions, which are well-constrained and can be efficiently extensible to high-dimensional problems. An illustration of this approach is shown below (Fig 1.). It can be observed that as the images come into alignment, the samples in the joint complex phase order distribution becomes tightly packed around a n-dimensional hyperline.

Figure 1 illustrates concept of joint complex phase order distributions

Figure 1

Examples of registration results using this approach are shown below (Fig 2.) for both medical and remote sensing imagery.

Example of registration result when using joint complex phase order distributions approachExample of registration result when using joint complex phase order distributions approach

Figure 2

Related people

Directors

Alexander WongDavid A. Clausi

Related research areas

Computer Vision

Image Segmentation/Classification