Study On The Image Registration Of Division Of Amplitude Polarimeter

With the development of polarization optical detection,It has become a hot topic to detect the target using polarization characteristics.During the period,various types of polarization detection devices have been developed and used in military target detection,atmospheric detection and many other fields.Among these devices,the division of amplitude polarimeter has the structure of multiple optical paths.If you want to get the polarization information of the target,the system must calculate the information by using multiple images that obtained at the same time.The precision of the image registration method is directly related to the detecting precision of the polarimeter.Therefore,it is very important to design a registration method,which have high registration accuracy and can register the images from multiple optical paths of polarimeter.In the detecting scene of polarimeter,the current image registration methods can not register images well under the condition that the target features are not obvious,the image features are difficult to extract,and the gray difference between the images are large.Based on the above situation,similarity measures function which is suitable for the images from mutiple optical paths of the polarimeter is proposed.Based on the function and combined with an optimization search algorithm,the image registration algorithm which can register images from mutiple optical paths of the polarimeter is completed.First,according to the phenomenon that the abnormal information in the polarization information image is caused by the position errors between mutiple images,The extraction algorithm of the similarity measure function based on the polarization information is proposed.In addition,according to the analysis of image characteristics,multi-parameters in the algorithm are adjusted so that the similarity measure function can measure registration degree between multi-images accurately.Then,according to the proposed similarity measure function,the workflow of the image registration of the polarimeter is designed to ensure that the similarity measure function just measures the position errors between reference and floating images.In addition,according to the structure of polarimeter and the image characteristics of the target,the form of position errors that exists between reference and floating images is analyzed.Subsequently,the appropriate geometric transformation is selected,meanwhile,the parameters of geometric transformation between reference and floating images and the technology of image resampling are determined to meet the requirements of the image registration algorithm.The genetic algorithm is selected as the optimization search algorithm for searching optimal parameters of geometric transformation.In the optimization process,the optimal geometric transformation parameters are obtained by taking the similarity measure function based on polarization information as the objective function and going through multi-generation iteration.The design of image registration algorithm is completed by integrating the above steps.Finally,The image registration algorithm is verified by using the simulated images and the real images acquired by the polarimeter.In addition,the accuracy of image registration is measured by mutual information(MI)between images.The experimental results show that mutual information of two simulated images registered by using image registration algorithm proposed in this paper is up to 2.6925,which is higher than the accuracy of commonly used image registration methods based on image features.Similarly,the mutual information of two real images registered by using image registration algorithm proposed in this paper is up to 1.8493,which is also higher than the accuracy of commonly used image registration methods based on image features.More accurate results of polarization information can be obtained by using images registered by the registration algorithm proposed in this paper,which proves that the accuracy of image registration algorithm proposed in this paper basically meets the image registration requirements of the polarimeter.