The Registration And Fusion Of Multi-sensor Images Based On Mutual Information

Multi-sensor images refer to the images produced by the sensors with differentimaging mechanism. Compared to single sensor image, the image information providedby multi-sensor images has reliability, complementary and redundancy. So people canhave a more accurate, more comprehensive and reliable image description to the scenegoal by the fusion image between multi-sensor images. A rapid and accurate registrationalgorithm can improve the accuracy of the fusion, also be an important prerequisite forimage fusion. So, a method based on mutual information and improved geneticalgorithm search strategy was proposed to realize the registration of the multi-sensorimages，and realize the fusion of images by wavelet transform.The primary work of the thesis is as follows:(1) Research mainly focused on the mutual information matching algorithm,including its principle and concept. On the basis of mutual information matching, thisthesis studied the effect of the spatial information in mutual information. In order toimprove the matching accuracy of mutual information in multi-sensor images, thisthesis focused on the gradient mutual information algorithm and regional mutualinformation algorithm. Verified by experiment, the recognition of the multi-sensorimages’ matching surface, and the peak of the optimal value was more apparent and lesscalculation by the regional mutual information algorithm.(2) The search strategy was a very important part in the image registration, decidedthe speed and accuracy of the registration algorithm. Therefore, global optimizationalgorithm-genetic algorithm was introduced in the image registration based on mutualinformation. Research mainly focused on the principle of genetic algorithm, and theimprovements in the premature and the slow convergence speed of the standard geneticalgorithm. Population diversity and relevance operation were introduced to improve thecrossover operator of genetic algorithm, also improve the global convergence of thealgorithm. And Powell algorithm was introduced to optimize the individual locallybefore the preservation of elite individual, to accelerate speed of the local convergence.(3) Finally, we needed to do the fusion for the images which had already been registered; this thesis focused on the application of wavelet transform in the imagefusion, and introduced the quality evaluation standard of the image fusion. Confirmedby experiment, the wavelet transform was better than the other algorithm in themulti-sensor images fusion whether by visual or quantitative evaluation.In this thesis, infrared image and visible light image, SAR image and optical imagewere used to do the registration and fusion experiments, to verify the feasibility of thealgorithm. The result proved that regional mutual information algorithm had betterrobustness and accuracy, and the improved genetic algorithm was able to maintain goodglobal convergence and faster convergence speed.