Research On Multisensor Image Registration

We can obtain much information from multisensor images, such as multispectrum, temporal changing, spatial magnitude and so on. It is helpful to give a more comprehensive description of external things through fusion of multisensor images. However, limited to the condition and situation of the given application, images acquired by different sensors may be misregistered. Besides, these images may also be acquired at different times. Therefore, images need to be registered beforehand, so image registration is a key technology in multisensor image processing.We proposed a novel approach in this paper to solve the problem of local extremum in image registration by introducing a decision-making process.sFirstly, a deciding model, which emploies multiple similary measures based on Dempster-Shafer evidence theory is built in this paper. By fusing information from multi-similarity measures using Dempster-Shafer evidence theory, a decision is figured out about the current outcome is global extremum or not. The decision mentioned here can be one of these three types: credible, unauthentic and suspect. Simulation results in this paper showed the effectiveness of the model.Then, we applied this model to a registration approach based on simulated annealing-simplex method(SMSA) optimazation technique. The decision educed from the model is used as a feedback guidance to improve the optimization technique. When the decision is credible, the current outcome is regarded as the global extremum and we speed up annealing process to enhance the convergence rate of the optimization process. When the decision is unauthentic, the current outcome is regarded as a local extremum and an"ascending"process is introduced into optimazation technique to increase the acceptance probability of every state, which can help optimazation process jumping out of the local extremum. When the decision is suspect, inerratic annealing is used as original SMSA algorithm. Experiments of this paper demonstrate that, our method is helpful for optimization process to avoid relapsing into local extremums, so as to achieve better reliability and higher precision of image registration algorithm.