Multi-sensor Image Registration Algorithm

Image registration is a fundamental problem in the field of image processing, and isalso the first problem to be solved in image fusion. It is widely used in medical imageanalysis, remote sensing image processing, object recognition and computer vision. In thepast few decades, many scholars have done a lot of research work in the field and made alot of achievements. But as the increasingly demanding of high-quality images andreal-time processing of massive data. The research about how to improve the accuracy andspeed of the image registration is particularly important. Aiming at these problems, in thedata support of a large number of infrared, visible and multispectral images, Analyzes thekey technologies in heterologous, homologous image registration. Now the main researchcontents are summarized as follows：（1）Analysis the development status of image registration techniques and point outthe shortcoming of the registration method at this stage, introduce the principles andprocesses briefly, introduce the existing registration methods and key technology,advantages and disadvantages of each type of methods.（2）For the gray-based registration method, the key technologies(i.e. similarityfunction and optimized search algorithms)are described in detail. Four kinds of similarityfunctions are analyzed in aspect of registration accuracy and speed; Three kinds of searchalgorithms are analyzed in aspect of advantages and disadvantages, the occasion and theimpact on registration results.（3）For the difficult problem of heterologous image registration, two effectiveregistration methods are proposed: First, for the images having a clear edge feature, thetraditional method which treating the mutual information as similarity function is improvedthe edge information of two images are used and angle difference in the edge direction asthe third argument is added to the mutual information, can effectively improve heterologousimage registration; Secondly, for the low correlation of heterologous images, presents animage registration method based on the edge and cross-correlation, removing thelow-correlation part and high correlation part is only be used for image registration. Bychoosing an appropriate similarity function and optimization algorithm, function not only can effectively avoid getting into local minimum, but also greatly improve the success rateof registration.（4）For the phase-based edge detection algorithm, proposes a feature extraction andmatching method, combine the Hausdorff distance and Gaussian function as the matchingcriteria, combine the matching points and avoid the impact of outliers on the registrationresults.Experimental results show that the proposed methods can effectively solve theproblems about accuracy and speed on image registration, can be used in the field ofmilitary and medicine.