Infrared And Visible Image Fusion Via Densenet

Image fusion technology in the remote sensing in real life,image enhancement,monitoring,detection plays an important role in many fields such as medicine,and the research direction in the infrared and visible light image fusion is a crucial technology branch,that can be widely applied to target recognition test,production safety monitoring,and military operations in all areas.This study briefly introduces the relevant technical theories in the research field,as well as the existing methods in this field.Among the existing image fusion algorithms,this study first proposes a new fusion framework called Dual-Transformer(DT)based on dense networks for infrared and visible image fusion.The fusion framework extracts sufficient detailed information from the source image through a dense network encoder with a jumper connection.On the other hand,DT is used to focus on different aspects of information and integrate all aspects of information to capture local and remote information.Through a large number of experiments on 18 pairs of test data fusion results,it is proved that the mean values of each evaluation index are higher than other comparison methods,thus proving that the proposed method has excellent fusion characteristics in the direction of infrared and visible image fusion.In addition to the first method to prove that dense networks have good fusion performance,a coupled dense network model based on conditional probability is proposed in the thesis.The model includes an autoencoder network and a fusion network,which can retain more characteristic information of input data the for thesis.Firstly,the infrared image and the visible image are input into the two networks respectively,and a good image is reconstructed by the autoencoder network,and the feature information is extracted by the auto-encoder network coupling.Finally,the objective function reconstruction of the conditional probability fusion image is carried out by the fusion network.In the thesis,a lot of static and video experiments are carried out to verify the model.Through quantitative and qualitative evaluation of fused images of 20 pairs of test data and 18 pairs of video sequences,the mean values of MIN,QABF,QY and VIFP are all higher than those of other methods,and the results of single image fusion are MIN=2.0346,QABF=0.6098,QY=0.9178 and VIFP=0.5293.The results show that this method can fuse infrared and visible light well and achieve better fusion performance than other methods.