Infrared And Visible Image Fusion Via Gradientlet Filter

Image fusion is a technique,which synthesizes multiple images obtained by different imaging equipment in the same scene or the same imaging equipment under different conditions into a composite image.Among them,infrared and visible image fusion is a hot topic.The significance is to integrate the redundant and complementary information of infrared and visible images into the final fusion image,so that the fusion image has targets that are more prominent and richer detailed texture than single input image,which is convenient for subsequent image processing tasks such: concealed weapon detection,target detection and tracking,video monitoring,etc.Since the original infrared image directly obtained from the sensor often has the problem of low contrast,it is necessary to enhance the infrared image before infrared visible light fusion.Traditional histogram equalization technology can meet the needs of low dynamic range infrared enhancement.However,with the development of science and technology,high-quality infrared sensors can capture infrared images with high width and wide dynamic range.At this time,it is not only necessary to enhance the infrared image,but also to compress the dynamic range of the image.If traditional histogram equalization technology is used,there will be the problem of missing details.However,several new infrared enhancement techniques based on multi-scale with high dynamic range will appear pseudo-contour in some scenes due to the edge blur or gradient inversion during decomposition.Multi-scale infrared visible fusion algorithms have similar problems.In addition,most of the fusion algorithms only design fusion rules from the perspective of texture information and ignore the brightness characteristics of the image,resulting in low contrast and unobtrusive target of the fusion image.In order to solve the above problems,this paper makes the following work:Firstly,in order to solve the problem of false contours caused by edge blur or gradient inversion and the shortcoming of neglecting brightness features only considering the details of texture features in traditional multi-scale methods.This paper proposes an image filtering algorithm-gradientlet filter from the perspective of brightness and gradient separation based on weighted least squares.The experimental results prove that the filter can not only effectively preserve the edge of the image to avoid the false contour phenomenon,but also separate the brightness feature and the gradient feature and contain multiple adjustable parameters with clear meaning,which has strong flexibility.Secondly,in order to enhance the contrast and detail of the image while compressing the high dynamic range infrared image and reduce the impact of noise and false contours.In this paper,a high dynamic range infrared enhancement technique based on gradientlet filter decomposition is proposed in the multi-scale framework.The algorithm uses gradientlet filter to decompose the image into approximate layer that saves the image brightness background and residual layer that contains image gradient texture and noise.After decomposition,a simple histogram projection is used to improve the image contrast of the approximate layer without noise and small details.Meanwhile,the adaptive double plateaus histogram equalization is used to improve the detail information of the residual layer and suppress the noise.Finally,the approximate layer and residual layer are reconstructed and remapped to obtain the final enhanced image.Experimental results show that the proposed algorithm is superior to the histogram method in preserving details and effectively reduces the influence of noise and pseudo-contour.Finally,for the current multi-scale fusion algorithms with low contrast,inconspicuous targets and false contours,this paper proposes a fusion algorithm based on gradientlet filters and image saliency mapping.The algorithm uses gradientlet filtering to separate the brightness and gradient features of the image.According to the different characteristics and visual characteristics of the approximation layers and the residual layers after decomposition,contrast and gradient saliency maps are proposed,and corresponding weight matrixes are constructed.Finally,the approximate layers and the residual layers are fused and reconstructed to obtain the final fused image.Subjective and objective experimental results on public data sets show that this method is superior to several other methods in maintaining image contrast,improving target saliency,preventing false contours,and reducing noise.