Fast Image Registration by LB Method

Abstract

Image registration is a key pre-procedure for high level image processing. However, for the complexity and accuracy of the algorithm, the image registration algorithm always has high time complexity. This means it is not suitable for realtime image processing, such as real-time target tracking. To speed up the registration algorithm, parallel computation is a good solution, for example, parallelizing the algorithm by LB method. In this paper, we proposed a LB based model for image registration. The main idea of our method consists in simulating the convection diffusion equation by establishing a LB model. Experiments shows our model is effective. Theoretically, our model can not get more accurate registration than the classical numerical method. But as a kind of numerical tool, our model is stable and faster, the most important is the potential for parallel image processing.