Research On The Fusion Algorithm Of Infrared And Visible Light Images

The goal of image fusion is to fuse the complementary information of multiple source images from the same scene to produce high quality composite images.Infrared image reflects the energy distribution of the target under infrared thermal radiation,which is not easily affected by complex conditions such as wind,sand,smoke,etc.,but its visibility is not very ideal,especially the performance of object texture details is poor.Visible light image is mainly related to the light reflection of the target scene,and the object identification is high,but it is easy to be affected by the external environment,especially when it is occluded,it cannot accurately capture the target feature information.Therefore,infrared and visible image fusion can integrate the advantages of the two kinds of imaging.By combining the complementary information of the two,the scene resolution and the ability to capture target features can be better improved.In order to improve the image quality of infrared and visible light fusion,this paper mainly studies multi-scale transformation,pulse coupled neural network,convolutional sparse representation and Laplace energy method.The main innovative work and contributions are as follows:(1)Due to the low integration degree of low-frequency information in nonsubsampled Shearlet Transform(NSST),Sparse Representation(SR)is easy to produce "pseudo Gibbs" effect,and complex parameter setting of Pulse Coupled Neural Network(PCNN)model,Based on Adaptive Sparse Representation(ASR)and Adaptive Pulse Coupled Neural Network(PA-PCNN)model,an infrared and visible image fusion algorithm in NSST domain(NSST-ASR-PAPCNN algorithm)is proposed in this paper.The algorithm first decomposes the source image into low frequency subbands and high frequency subbands through NSST.Then,the adaptive sparse representation(ASR)model was used to fuse the sparse coefficients in the low-frequency part of NSST domain.At the same time,the parameter adaptive pulse coupled neural network(PA-PCNN)model is used to fuse the corresponding high frequency parts.Finally,NSST inverse transform was performed on the fused low frequency and high frequency bands,and the fusion result was reconstructed.Experimental results show that the proposed algorithm solves the "block effect" problem of traditional SR model,overcomes the difficulty of setting free parameters in PCNN model,and is superior to existing algorithms in subjective vision and objective evaluation.(2)according to the sampling profile under the wavelet Transform(Non-subsampled Contourlet Transform,NSCT)Sparse sexual weak low frequency part,traditional Sparse Representation of the "sliding window" operation damage link between image fusion results feature information loss serious shortcomings,is proposed in this paper,based on convolution Sparse Representation(Convolutional Sparse Representation,CSR)and energy characteristics of NSCT domain of infrared and visible light image fusion algorithm(referred to as: NSCT-CSR-LLE algorithm).The algorithm firstly NSCT decomposition to source images,and low frequency subband NSCT domain by gaussian filter is decomposed into low frequency base component and the detail characteristics of the component,the corresponding part,USES the Local Laplace Energy method(Local Laplace Energy,LLE)and convolution sparse representation of fusion,coupled with low frequency subband image fusion,in the high frequency part,according to the low-level visual features by building a new high frequency subband activity measurement method to fusion,finally,the low frequency and high frequency subband fusion image reconstruct NSCT inverse transform to get the final fusion image.Experimental results show that this algorithm solves the pseudo-Gibbs effect of the sparse representation model,extracts the edge details of the source image effectively,and improves the overall visual effect of the image.