Fast, Robust Image Registration for Compositing High Dynamic Range Photographs from Hand-Held Exposures

Abstract In this paper, we present a fast, robust, and completely automatic method for translational alignment of hand-held photographs. The technique employs percentile threshold bitmaps to accelerate image operations and avoid problems with the varying exposure levels used in high dynamic range (HDR) photography. An image pyramid is constructed from grayscale versions of each exposure, and these are converted to bitmaps which are then aligned horizontally and vertically using inexpensive shift and difference operations over each image. The cost of the algorithm is linear with respect to the number of pixels and effectively independent of the maximum translation. A three million pixel exposure can be aligned in a fraction of a second on a contemporary microprocessor using this technique.