The analysis of image time series requires a correlation of the information between two images. The gradient flow registration is a method for correlating this information by successively minimizing an appropriate energy along its gradient. A graphics hardware implementation of this approach to image registration is presented. The gradient flow formulation makes use of a robust multi-scale regularization, an efficient multi-grid solver and an effective time-step control. The locality of the involved operations implies a data-flow which is very well suited for an acceleration in the streaming architecture of the DX9 graphics hardware. Therefore, the implementation obtains registration results at very high performance, registering two $256^2$ in less than 2 seconds, such that it could be used as an interactive tool in medical image analysis.
@Article{StDrRu03regSimple,
author = {Robert Strzodka and Marc Droske and Martin Rumpf},
title = {Fast Image Registration in {DX9} Graphics Hardware},
journal = {Journal of Medical Informatics and Technologies},
year = {2003},
volume = {6},
pages = {43--49},
month = {Nov},
}