Research On Fusion Method Of Infrared And Visible Image

Image fusion technology is now a popular area of information fusion research,it is the process of extracting the favorable components of the source images by extracting the information about the same target observed by multiple source sensors through methods such as image processing and computer technology to obtain a high quality fused image with the most optimal technique.And the infrared and visible image fusion studied in this thesis is an important branch of observer image fusion,which has important applications in military defense and civil surveillance.Infrared sensor imaging by detecting the thermal radiation mechanism of the object,has the advantages of unconditional observation,anti-interference,and disadvantages such as low contrast and poor visibility.Visible light sensor by accepting the object light source reflection mechanism imaging,with high resolution,good visual perception and other advantages,and vulnerable to external influence,can not observe the hidden target and other disadvantages.Therefore,the use of the two imaging complementary characteristics,effective fusion processing of infrared and visible images is very important research technology.Firstly,this thesis gives an overview of the background significance of infrared and visible image fusion,introduces the status of important theories in research and applications at home and abroad,and organizes the main contents and chapter structure of this thesis.The definition of image fusion and the mechanism of infrared and visible imaging are organized,and the necessity of image fusion work is demonstrated by the difference of characteristics between them.After that,the image pre-processing part,the system structure of image fusion and the important indexes to evaluate the image fusion quality are introduced immediately,and the algorithms of image fusion methods with different multi-scale decomposition are introduced.The differences and variations of pyramid transform,wavelet transform,Non-Aubsampled Contourlet Transform(NSCT),and Non-Subsampled Shearlet Transform(NSST)algorithms are analytically studied.Secondly,the infrared and visible image fusion algorithm based on Retinex-enhanced multiscale decomposition is investigated.To solve the problems of poor contrast of fusion results,blurred target margins,and loss of background detail information under low illumination conditions of traditional I R and visible image fusion algorithms,the algorithm first uses Retinex to perform SSR algorithm information enhancement processing on the weak visible image,performs multi-scale decomposition on the source image using cross-bilateral filtering to obtain the base layer and detail layer image information of the source image,adopts a combination of absolute value taking larger strategy and bootstrap filtering for the base layer image The fusion method combining absolute value enlargement strategy and guided filtering is used for the base layer image,and the fusion method of constructing weight map and significant map is used for the detail layer image,and finally the processed base layer and detail layer images are fused by using weighted fusion to obtain the final fused image.Analyzed from both subjective and objective levels,compared with other algorithms,the infrared and visible image fusion algorithm based on Retinex-enhanced multiscale decomposition has better visual effects,with high contrast,clear target contours,rich information retention,etc.,and many indicators have good achievements.Finally,a new fusion method is introduced,the infrared and visible image fusion method based on NSST and improved PCNN.The principle of this method is to first decompose the infrared and visible images using NSST,and obtain the low-frequency and high-frequency components about the images after the decomposition process,after which the information of the low-frequency components is fused using an improved spatial frequency-based impulse coupling neural network fusion criterion,and for high-frequency information fusion,a local variance adaptive weight assignment strategy is used.Finally,the information parameters processed separately are inverted by NSST to obtain the final fused image.The method proposes an improved method in this thesis by organically combining the low-frequency component to maintain the basic content of the image and the high-frequency component to protect the image detail information,and the experimental results verify the feasibility and superiority of the fusion performance of the method in the image fusion problem.