Research On Fusion Of Infrared And Visible Light Image Based On Rolling Guidance Shearlet Transform

Infrared and visible image fusion is the most widely used and valuable information fusion method.Because of the good complementary characteristics between infrared and visible images,the effective combination of these two spectra can improve the recognition of infrared targets and obtain more detailed and accurate visual saliency information,so that we can judge the location of the heat source more accurately in the harsh environment.At present,multi-scale transform algorithms are a type of method that can effectively solve the problem of fusion of infrared and visible light.The core idea is to decompose the image into a series of sub-band components with different characteristics,and then use certain fusion rules to process them separately.Finally,a fused image will be obtained.However,the existing multi-scale transform algorithms all have poor ability to capture details,which is easy to miss the edge detail information in the image,and they do not have adaptive decomposition characteristics.In addition,the existing fusion rules can't well eliminate the spectral differences between the source images,and there are no more effective ways to improve the image quality of fusion image when its target is in the environment of low brightness or occlusion.In view of the above problems,in order to effectively improve the visual performance of the fusion algorithm,the research content and main innovations of this article are as follows:1.A new multi-scale transform tool is proposed:rolling guidance shearlet transform?RGST?.RGST can effectively combine the scale sensitivity of rolling guidance filter with the directional characteristics of shearlet filter,which can be adaptively decomposed according to the scale characteristics of image edge,and the multi-scale and multi-directional component information after decomposition is more precise.In addition,there are no up sampling and down sampling operations in the process of decomposition and reconstruction of RGST,and it does not need to meet any constraints.It is only a simple linear difference and superposition calculation,so it has translation invariance and high calculation efficiency.2.For the fusion of low-light-level and infrared image,an algorithm based on RGST and variational optimization model is proposed.Firstly,RGST is used as the multi-scale transformation tool to obtain the approximate layer and detail layer component of the source image.Secondly,a variational optimization model based on?₂-energy minimization is used as the fusion rule for the approximate layer components which reflect the image energy characteristics.This method not only constrains the fusion image to have the same pixel intensity as the given infrared image,but also effectively combines the spectral saliency information in the image,so it improves the visual observation of the fusion image,and makes up for the brightness defect in weak light.In addition,the variational model based on hybrid?₁-gradient regularization is used to guide the fusion of detail components,so that the fused image has more clear edge details.The final experimental results show that this algorithm has better visual effect and is superior to other existing typical methods in the low-light-level environment.3.In order to effectively combine the thermal imaging characteristics of infrared sensor,and make the target in visible image can be fused and identified in the background of occlusion and smoke,an image fusion algorithm based on RGST and adaptive dual-channel pulse coupled neural network?PCNN?is proposed.The new algorithm still uses RGST as the multi-scale analysis tool,and then uses the adaptive dual-channel PCNN model as the fusion rule of each image component.For the PCNN model of approximate layer component,the visual stimulus factor?VSF?composed of the salient features of image spectrum is used as its external stimulus;for the PCNN model of detail layer component,the improved spatial frequency operator?ISF?is used to activate the pixel neurons in the model.In addition,the salient structure factor?SSF?and the gradient singular value operator?GSV?obtained based on the matrix singular value decomposition are used as the adaptive linking strength of the above two kinds of component PCNN models.Through a number of experiments,it is shown that this algorithm can solve the problem of fusion imaging in which the target is under cover more pertinently.The obtained image has higher contrast,more salient thermal infrared visual features,and more prominent target information.