Sonar Image Registration And Fusion Technology Research

In underwater acoustic detection,sonar image visually characterizes the target echo data,while its registration and fusion play a critical role in later image processing and analysis.Due to the principle of acoustic imaging,leading sonar image has some unique attributes,so there are difficulties for sonar image registration and fusion,simultaneously,certain particular conditions or demands for process.In this sense,research for sonar image processing method still has broad application prospects and urgent practical needs,although there have been a large number of reliable image registration and fusion algorithms in the field of optics,remote sensing,medicine and other.In this paper,some significative methods for sonar image registration and fusion were discussed.For the sonar image registration,more popular mutual information was chosen as the theoretical foundation.The main research of the rigid image registration was similarity metric function and optimization methods.One of which took pixel space contact into account,called metric function,and the other avoid metric function falling into local extreme,called optimization methods.Among these methods,some of them were very mature,or some had been proposed for many years,but were rarely applied in sonar image registration.That is why in this paper,by a lot of simulation results,the differences in performance of these methods in sonar image registration were discussed respectively.And under the image registration based on mutual information framework,specific expressions of similarity metric function based on gradient mutual information were proposed,and hybrid optimization strategy based on PSO and Powell method were established,then using the multi-resolution analysis to improve the sonar image registration performance.After summing up the sonar image registration algorithm,some representative sonar images were conducted registration.By comparing other rigid deformation registration algorithms,our method was excellent performance and robust stability.Subsequently,on the basis of rigid deformation registration algorithms proposed,we tried the registration of non-rigid deformation sonar image.We analyzed the inadequate of the Demons algorithm in the sonar image registration to construct a Demons model based on gradient mutual information,and proposed the corresponding sonar image registration algorithm.Meanwhile combining the characteristics of sonar image,the important factors in Demons algorithm were analyzed,and some valuable conclusions were obtained.Applying Demons model based on gradient mutual information in registration of non-rigid deformation sonar image,by comparing traditional Demons model and active Demons model registration,experiment results confirmed the effectiveness of our model.For the sonar image fusion,this paper started the research from the fusion and evaluation.Firstly discussed the application of general fusion framework based on Piella theory in sonar image fusion,and multi-scale image fusion based on Piella framework were introduced,including multi-resolution analysis,active measure,matching measure,decision module,as well as synthesis module.According to the characteristics of sonar image and processing,We expanded Piella multiscale image fusion framework,abandoned the match measure and decision module,and proposed using last level information to complete the fusion of current level approximation image,corresponding image quality assessment parameters were put into establishing the rules of detail image.Simulation results proved that this progress to the sonar image fusion had great significance,by which more conducive to improving the quality of fusion.Secondly,this paper systematically studied and analysed a variety of subjective and objective methods of image fusion evaluation criteria,and summarized the general principle suitable sonar image fusion evaluation.On the basis of principle,we intensively analysed some classical sonar images fusion method,and obtained some corresponding conclusions.Finally,this paper proposed a sonar image fusion evaluation methods based on regional similarity,simulation experiment results showed that it was consistent with the human visual system.