Research On The Fusion Method Of Infrared Polarization And Intensity Images Based On Difference Features

Fusion of infrared polarization and infrared intensity images is the front key technologyin the thermal infrared target tracking and recognition, it has been widely applied in themilitary and civilian fields. Currently, fusion algorithms of infrared polarization and intensityimages are mostly pre-determined, but their difference features are dynamic change in theactual application, pre-determined fusion algorithm obviously can’t change along with thechange of difference features, this often leads to poor adaptation and failure of fusionalgorithms.In order to improve the deficiency of the original fusion algorithms, this paperestablishes a mapping relationship between difference features and fusion algorithms, themapping relationship is used to realize the driven fusion of difference features. The mainresults of this research are as follows:(1) This paper obtained the main difference features of images based on analyzing themain difference characteristics between infrared polarization and infrared intensity, and putforward some representation and extraction methods of difference features. From the maindifference features selected some low correlation and highly complementary differencefeatures to build a difference feature set. From the existing typical fusion algorithms selectedthe good effect and fast algorithms to build a fusion algorithm set.(2) The possibility distribution of fusion algorithms which were effective in a certaindifference feature was constructed by the possibility theory, the elements of a certain cut set ofthis distribution chose acted as fusion algorithms which are driven by the difference feature,namely, it reduced the dimension of fusion algorithm set to obtain a new fusion algorithm set.The source images and the fused images were divided into blocks, difference feature valueswere calculated for each pair of sample block of source images, and evaluated the fusionimage sample block of each algorithm, and then got the optimum algorithm number of each sample block, and counted the probability of the algorithm number. The correspondingrelationship between difference feature value and the probability of fusion algorithm numberwas established, thus the mapping relationship between difference features and fusionalgorithms was established.(3) Using the possibility distribution of difference features which described thedifference of source images to select the elements of a certain cut set which acted as drivingsource image fusion difference features, and the difference features substituted into themapping relationship between difference features and fusion algorithms, thus determined theweight of each fusion algorithm driven by difference feature, then used algorithms driven tointegrate source images respectively, and the fusion results were weighted fusion to achievethe driving fusion of difference features.Using the sixteen algorithms and the proposed method conducted fusion experiments tointegrate test images respectively, these fusion results obtained were evaluated by thesubjective and objective indicators, comparative analysis of results were that the proposedmethod had more advantages in the fusion effect and performance, and effectively solved theproblem that fusion algorithm did not vary with the change of the difference feature.