Research On Multilevel Mapping Between Difference-Features And Fusion Strategy For Dual-Modal Infrared Image Fusion

Dual-modal infrared intensity and polarization images have great fusion value because of their unique imaging advantages and rich complementary information.Therefore,how to configure the optimal fusion strategy(including fusion algorithm and fusion rule)for the fusion of dual-modal infrared images to achieve the high-quality fusion is the key to fully use the complementary information of dual-modal infrared image,and is also an important prerequisite for the efficient implementation of various subsequent image tasks,such as target recognition,target tracking and object classification.However,in the dual-modal infrared image fusion,the configuration or selection of fusion strategy are mostly based on previous knowledge and relatively fixed.As we all know,in practical application,the imaging scene is not only complex and changeable but also unpredictable,which directly leads to the fact that fusion strategies selected based on experience are difficult to obtain high-quality fusion images in most cases.In order to make the fusion strategy adapt to the fusion scene adaptively,this paper uses the difference-features to quantitatively describe the difference complementary information between the two types of images based on the function and attribute of the difference-features,and establishes the multilevel mapping relationship between the difference-features and the optimal fusion strategy under the multiscale decomposition framework,so as to realize the selection of the best fusion strategy driven by the difference-features of dual-modal infrared images.Moreover,because the global principal difference-feature is only the overall average reflection of complementary information,it is difficult to accurately describe the local differences.Therefore,this paper further proposes a block fusion method based on difference-feature measurement,to ensure the global optimality of the fusion results.To sum up,this paper studies the dual-modal infrared image block adaptive fusion method based on depth set-valued mapping under the framework of multi-scale decomposition,and the main innovative work is as follows:(1)The construction of mapping set of multilevel mapping between difference-features and fusion strategy: Due to the lack of in-depth exploration of the nature and attributes of the difference-features,the difference features constructed are lack of standardization.So,the constructed difference-features cannot describe the difference complementary information of the two types of images comprehensively and effectively quantitatively.In order to solve this problem,starting from the role of difference-features,this paper analyzes and gives three essential attributes of difference-features for the first time,namely,describing information difference,representing difference independence and guiding fusion effectiveness.At the same time,according to these three basic attributes,combined with the principle of dual-mode infrared imaging and their image characteristics,six difference-features are constructed to quantitatively describe the difference complementary information of gray-difference,edge-difference and texture-difference of the two kinds of images.And then the rationality and effectiveness of the above-mentioned six difference-features are verified by experiments in turn.As for the fusion strategy set,under the framework of multi-scale decomposition,a large number of typical fusion algorithms and rules are selected,and the rationality of algorithm parameters and combination of high-and low-frequency rules are compared and analyzed for the fusion of dual-mode infrared images.Finally,a fusion strategy set suitable for dual-mode infrared image fusion is constructed.(2)Research and construction of fusion-effectiveness between the difference-features and fusion strategies of dual-mode infrared images: In order to make the fusion-effectiveness reflect the fusion effect of difference-features and the objective fusion quality of fusion results,so as to link the difference-features with the optimal fusion strategy.This paper analyzes and clarifies the difference between the fusion degree of difference-features and the fusion-effectiveness.This paper constructs a more reasonable difference-feature fusion degree by comparing the advantages and disadvantages of different-feature fusion degree.In addition,by analyzing the function and connotation of fusion-effectiveness,a new fusion effectiveness based on the correlation between difference-features fusion degree and objective evaluation metrics is proposed for the first time.The proposed fusion-effectiveness not only takes into account the objective evaluation of fusion quality,but also comprehensively describes the relationship between fusion strategy and fusion quality.(3)The establishment of multi-level mapping relationship between the difference-features and the optimal fusion strategy and its fusion form: In order to overcome the problem that the selected strategy is fixed and difficult to be optimal due to the selection of fusion strategy by experience.This paper,firstly,establish the shallow multi-level mapping relationship between the difference-features and the optimal fusion algorithm and the optimal rules(including high-frequency rules and low-frequency rules)and further reveals the influence of the difference degree amplitude of the difference-features on the above mapping relations.Finally,the deep multi-level mapping relationship between the difference-features and the optimal fusion strategy is established.Then,the best fusion strategy can be adaptively selected and configured by using the difference-features,which paves the way for selecting the optimal fusion strategy adaptively according to the difference-features and their amplitudes.At the same time,this paper further studies the application of the multi-level mapping relationship in the dual-modal infrared image fusion,and the implementation of deep fusion under the multi-scale decomposition framework,which makes the overall quality of the fusion of the two types of images significantly improved.(4)The realization of block adaptive fusion method of dual-modal infrared image based on multi-level mapping: In order to break through the limitation that all local regions cannot be the optimal fusion when the fusion is guided according to the global difference characteristics.This paper starts with the relationship between the global principal difference-feature and the local principal difference-feature,and proposes a block specific fusion method based on the local principal difference-feature.Based on the idea of region growing,a block aggregation segmentation method is proposed to segment the source image,and then the optimal fusion strategy is adaptively selected according to the main difference-feature of each block to implement block fusion.Finally,the fusion results of each block are stitched to form a complete fused image.Because each block is the optimal fusion in the fusion process,the final fusion result truly realizes the global optimization.