Research On Feature-level Image Fusion Method Based On Infrared And Visual Image

Fusion of infrared and visible images is an important subdivision in image fusionsector, which can sufficiently utilize complementary information from different sensorsto gain more detailed description of the image object and increase its collectedinformation accordingly. Comparing with decision-level fusion and pixels-level fusion,feature-level fusion can not only effectively eliminates redundant information ofmulti-feature relativity brought from subjective and objective factors, but also to themaximum degree retains effective identification information of multi-featureparticipated in the fusion. Due to very limited available feature-level fusion method andfurther popularization on its application sector, this content is still need to be verifiedthrough deep theoretical analysis and meticulous experimental tests. Just in thisbackground, this paper makes a deep study on feature-level fusion method of infraredand visible image.According to the characteristics of infrared and visible images and on the basis ofimage preprocessing and object feature extraction, this paper makes a study onfeature-level fusion method of infrared and visible images. The main study content ofthis paper is as follows:(1) Give an analysis of image features which are commonly used in featureextraction. As there is no socially accepted evaluation method on feature-level fusioneffects, the paper puts forth an evaluation criterion of integrated algorithm performanceand fusion feature recognition effects and introduces object identification methods ofSupport Vector Machines (SVM). In the meantime, this paper gives feature-level fusionflow diagram, and expounds the detailed fusion process.(2) Have a study on Covariance Matrix Algorithm, and according to CovarianceMatrix characteristics, put forth a covariance matrix feature-level fusion method. Basedon the extracted infrared and visible images characteristics, by structured regioncovariance matrix to conduct image feature-level fusion and make sure that featuresafter fusion has a strong rotation invariance, scale invariance and robustness, and has good discrimination ability on different objects.(3) Have a study on feature-level fusion method based on Principle ComponentAnalysis (PCA). On the basis of theoretical study, using Principle Components Analysismethod to make fusion on extracted image feature and structure the PCA fusioncharacteristics, and then to give algorithm process and carry on the experimentalanalysis.(4) Have a study on feature-level fusion method based on the improved immunegenetics. Aiming at the deficiency that conventional Immune Genetic Algorithm is easyto fall into "premature", this paper puts forth an Improved Immune Genetic Algorithmusing self-adoption strategy in which "elite keep" strategy is added, to ensure that theImproved Immune Genetic Algorithm can converge to the optimal solution as soon aspossible. Based on this research, also set forth the feature-level fusion method onImproved Immune Genetic Algorithm, and give a detailed analysis of above methodaccording to the algorithm itself performance and different objects recognition effects.