Efficient Image Alignment with Outlier Rejection

Image alignment is one of the most widely used techniques in computer vision. Applications range from optical flow and tracking to layered motion, mosaic-ing, and face coding. Of particular concern in many applications is the efficiency of the algorithm. This concern has led to the development of several efficient algorithms such as the Hager-Belhumeur algorithm and the inverse compositional algorithm. Another concern is the robustness of the algorithm. Image alignment is a form of template matching. What happens if the template does not match the image it is being matched to because the image is occluded? Similarly, what happens if specularities cause the brightness constancy assumption to be invalid? A solution to these problems is to use a robust form of image alignment. Most robust image alignment algorithms are inefficient however. In this paper we present a robust extension to the inverse compositional algorithm that is almost as efficient.