Study Of Multi-sensor Images Registration Based On Mutual Information

Image registration is the matching of image on space,which is received from the sensors of different time, scenarios and angles.Registration is one of the key technology of image processing.Image registration is widely used in fields like remote sensing,medical imaging,computer vision and pattern recognition and so on.Multi-sensor images refer to the two or more images with different imaging mechanism.Contrasted with the single-sensor images, multi-sensor images can get more comprehensive and accurate information with reliability, complementary and redundancy. With the deepening of the application, efficiency and accuracy of the image registration is gradually put forward higher requirements.Mutual information algorithm basing on the image entropy has better image effect,which is a kind of registration algorithm for the gray-scale statistical properties of the image without preprocessing of image segmentation.Given the large amount of calculation of the mutual information algorithm,to improve the efficiency of the image registration,this paper applied registration strategies from coarse to fine toward images with the method of wavelet decomposition.On the other hand, the mutual information algorithm usually will have problems without obvious peak point, wrong matching etc.when the two images are in higher gray contrast, louder noise and more serious impact.Therefore this paper applies the image’s space and gray information into registration measurement with the method of a new kind of measurement which combines the gradient space information with mutual information together.Through the experiment, it can be concluded that the methods above can contribute to the reduction of the calculation amount and the gray contrast influence, the measure curve sharp,easier registration parameters accuracy and reduction of the external factors influence,which proves the higher efficiency and accuracy of the methods demonstrated in this paper.