Research On Thermal Infrared And Visible Image Fusion

Image fusion is both an engineering science and an art. As an important branch ofImage Engineering, Image fusion has caused profound technological change during thepast30years. Image fusion rules and methods have been to imitate the cognitivefunction of our brains and visual perception system. Image fusion is gradually tobecome more and more intelligent and abstract in visual understanding.Thermal infrared and visible image fusion, widely required and applied in thefields of military surveillance, security monitoring and so on, has been a hotspot inimage fusion. On the basis of thermal infrared and visible images, according to thelimitations of region based image fusion frameworks and the enlightenment bycognitive image fusion, this thesis present a cognitive content based image fusionframework. The main work can be summed up as follows:(1) Four popular multi-resolution analysis algorithms Mallat, à trous, Contourletand NSCT were reviewed respectively. Taking into account the four algorithms’correlation in image decomposition and reconstruction, based on Mallat algorithm, thescale selection for thermal infrared and visible image fusion was studied. The resultsshowed that, based on an overall consideration of the spatial quality and the fidelity ofthe fused image, as well as the real-time performance and computation of the fusionprocess and other factors, three or four scale decompositions for thermal infrared andvisible images were chosen to get a better image fusion quality. On the basis of that, theregion based thermal infrared and visible image fusion was further studied.(2) The pixels, window and region based image fusion rules were reviewed, thelimitations of the region based image fusion frameworks were mentioned. We studiedand analyzed cognitive image fusion, inspired by this idea, and according to the modesof image understanding and cognition, a cognitive content based image fusionframework was proposed. Finally, this framework model was applied to the thermalinfrared and visible image fusion. The experimental results showed that this frameworkoutperformed the region based image fusion framework. This effect benefits from thatdifferent image fusion algorithms and rules in spatial or frequency domain could be simultaneously selected for different image contents by this proposed framework, whichcan maximize image fusion advantages.(3) On the basis of cognitive content based image fusion，thermal infrared-visiblecolor image fusion based on cognitive content and color transfer, and thermalinfrared-visible color image fusion based on cognitive content and thermal infraredimage inverse, were studied respectively. The experimental results showed that the colorof the first algorithm’s color fusion image was natural, vivid and close to that of realscene because of color transfer. According to the reverse polarity of infrared image, thesecond algorithm’s color fusion image was in line with human’s vision associations ofwarm red and cool blue, which efficiently improved the heat targets detection andidentification.