| With the rapid development of modern technology,fusion technology of visible light and infrared images has been increasingly applied in engineering and scientific research.Image fusion not only solves the problems of background blur and detail loss in infrared sensors,but also overcomes the dependence of visible light sensors on environmental conditions,and can construct images with both thermal radiation sources and multispectral information.Visible light images under low illumination often suffer from problems such as low overall brightness,high noise,and unclear details in dark areas,which affect the acquisition of useful information from the images.Therefore,this article enhances the visible light images under low illumination before image fusion,so that the fused image can obtain more useful information and have a wider range of applications.This article mainly studies three parts: enhancement of visible light images under low illumination,registration of heterogeneous infrared and visible light images,and fusion of infrared and visible light images.The specific contents are as follows:(1)In order to solve the problem that the existing image enhancement algorithms are easy to produce halo artifacts and lose details,this paper proposes a non-uniform weak light image enhancement algorithm based on light-dark segmentation,which can effectively enhance the dark areas while suppressing the halo in the bright areas.The algorithm extracts the brightness Y component of the original image through color space transformation,divides the brightness channel into bright areas and dark areas according to the average value,enhances and merges the bright and dark areas respectively,and processes the merged brightness image in gradient domain to solve the problem of edge fusion,and finally obtains the final brightness component through Poisson reconstruction,and finally obtains the enhanced image through linear color restoration.Combined with the subjective and objective evaluation results,it shows that the algorithm has obvious enhancement effect on uneven low-light images,and can effectively improve the enhancement deficiency and halo artifacts existing in the enhancement process of uneven lowlight images.(2)Different imaging devices have different spatial positions when shooting the same target,which makes the registration of multi-source images a necessary step of image fusion.In this paper,a registration algorithm based on SURF is used to complete the establishment of image fusion database.Firstly,the SURF algorithm is used to detect the feature points of infrared and visible light edge images,and the detected feature points are roughly matched,and the matching results are judged according to the minimum Euclidean distance.Next,the feature points are accurately registered through slope consistency;Finally,the registration of the two images is realized by affine transformation and bilinear interpolation,so as to construct a fusion database of visible and infrared images.(3)In the color fusion of visible light and infrared image,linear color transfer method based on increasing transfer weight is adopted to keep the natural color information of visible light image to the greatest extent and make the target more prominent.In addition,this paper adopts the dual-light image fusion method based on dual-decider in GAN network,and takes the visual saliency maps of two source images as the input of the decider to guide the generator to generate the fused image,so that the fused image retains more salient features of the original image.In addition,the gradient constraint is added to the loss function of the network to preserve the significant contrast and enrich the texture,and to prevent the background texture pollution caused by the edge diffusion of the high contrast target area.The experimental results show that the fused image has high definition and rich texture details,and it is superior to the existing methods in preserving significant contrast and enriching texture. |
