Research On Multi-spectral Images Registration

Images of multi-spectral are very different in grey level and image features, so we bring forward a new algorithm based on mutual information to match multi-spectral images. Considering of images with multi similar objects, we adopt normal mutual information to improve the robust of image registration. We use rigid body affine transformation to solve the problem with scale, rotation, and translation changes between two multi-spectral images, and use PV interpolation algorithm to solve the problem of local extremum. The compute-cost of mutual information is too large, so we bring forward a new algorithm based on wavelet and optimization. This algorithm speed registration process through shrinking the search space and reducing search times.Experimental results show that our algorithms solve the problem with multi-spectral images registration, and make registration time one magnitude faster than traditional algorithms. Simultaneously, the registration results achieve a high accuracy and stability.