Study On Techniques Of Multi-Sensor Image Registration

Image registration is used to match two or more images at different time, from different viewpoints, from different sensors accurately. Now the technology has been widely used in the areas of military affairs, remote sensing, medicine, computer vision, and so on.The paper mainly concerns research on the techniques of Multi-sensor image registration, proposes An Adaptive Genetic Algorithm (AGA) which can solve premature convergence problem and an image registration method based on wavelet multi-scale, the accuracy and speed of registration are improved effectively.An Adaptive Genetic Algorithm (AGA) is proposed to overcome easy to fail into premature convergence of the Standard Genetic Algorithm. In this method, the definitions of crossover probability and mutation probability are improved to change by individual fitness value in the evolutionary process. Adopting AGA to optimization process of image registration, Mutual Information and Alignment Metric are used as registration criterion respectively, the paper presents a registration method based on Mutual Information and AGA and a method based on Alignment Metric and AGA. Experiment results demonstrate these two methods have high accuracy and convergence speed.At present, the application of wavelet transformation in image registration domain is a research hotspot. The paper proposes an image registration method based on Wavelet multi-scale. In the method, on high-level, Partitioned Intensity Uniformity registration criterion is used to achieve rough registration for its high-speed and strong anti-noise, on low-level, using Mutual Information to complete refined registration; also adopting pruning principle of artificial intelligence to reduce search space. Experiments testified this method has high accuracy and reliability.