Research On Fusion Method Of Ultraviolet And Visible Images

In the technology of ultraviolet imaging,the target is imaged through a dual optical path system,In one of the path,the ultraviolet camera captures the ultraviolet light images by detecting 240~280 nm band,In the other path,the visible light camera captures the visible light images by detecting visible light band,and then,fuse the two images from the dual optical path systemand output the used images in one path.The imaging technology has got great advantages in corona detection and fault location in the fields of power and railway,and received strong support and promotion from the country.However,the precision of the image fusion is not high enough,and the definition is not good enough either,which bring so much inconvenience for UV detection,for this issue,a series of study is expanded in this thesis.In this thesis,the fusion of ultraviolet light and visible light in ultraviolet imaging is studied in depth,including image denoising,defogging,registration and fusion.In this thesis,the image is denoised first,and the corresponding defogging algorithm is used to improve the image quality.The accuracy of image registration directly affects the accuracy of target location.Therefore,the image registration algorithm is focused in the thesis to improve the the accuracy of registration.Image fusion is the focus of this research,it is to fuse the two images which contain completely different information into one image,keeping the information from the two original images maximally as the basis for the target judgment and positioning.It is of great importance for ultraviolet imaging to improve the quality of the image fusion.In this thesis,the Wiener filtering algorithm is used to remove the multiplicative noise and the salt and pepper noise,and the median filtering algorithm is used to filter out the Gaussian noise.The He algorithm is used to dehaze the image,and the visual effect is obviously improved,and the He algorithm is superior to other algorithms in terms of common indicators such as entropy and average gradient.In this thesis,the traditional registration algorithms are studied and the ICP-SURF(ICP,Iterative Closest Points.SURF,Speeded Up Robust Feature)algorithm is put forward,the registration result of this algorithm is superior to traditional SIFT(Scale-invariant feature transform),SURF and other algorithms in both subjective and objective evaluation indicators.In terms of image fusion,the NSST(Non-subsampled Shearlet Transform)algorithm is optimized and improved in the thesis.Firstly,the two images are decomposed by NSST,and the corresponding high and low frequency images of both images are obtained respectively.And then the PCNN(Pulse Coupled NeuralNetwork)algorithm is used to fuse the high frequency images,and the Adaptive Fuzzy Weighted Average Algorithm Based on Fuzzy Logic is used to fuse low frequency images.Finally,the NSST inverse transform is used on the fused high-frequency and low-frequency images,obtaining the final fused image.The fusion result shows that the algorithm proposed in this thesis is superior to shear wave,wavelet and NSCT(Non-subsampled Contourlet Transform)in both subjective and objective evaluation indicators,achieving satisfactory results.