Research On Image Fusion Algorithm Based On Rolling Guidance Filter And Sparse Representation

With the rapid development of sensor imaging technology in recent years,the image quality of sensors has been continuously improved.Different imaging sensors have different points of interest for target features.In order to obtain a more comprehensive representation of target features,multi-sensor image fusion has been rapidly developed.Multi-sensor image fusion can overcome the shortcomings of information provided by a single sensor.By complementing the superior information of a single sensor,a high-quality image with richer information and more comprehensive content can be obtained by fusion.The multi-scale transformation can extract the structural information in the source image well,and the sparse representation can obtain the sparse dimensionality reduction representation of the source image.Based on this,this paper combines the characteristics of multi-scale transformation and sparse representation to study the image fusion of multi-sensor.The main work of the thesis is as follows:1.The paper briefly introduces the background,significance and research status of image fusion,as well as transform domain algorithms such as multi-scale transform algorithm and sparse representation algorithm for fusion at the pixel level.The related theoretical knowledge of Rolling guidance filter and sparse representation is introduced in detail,and the methods of image quality evaluation and the principles of several commonly used objective evaluation indicators are given.2.Based on the multi-scale Guidance filter image fusion,a multi-scale Rolling guidance filter image fusion algorithm is proposed.Iteratively uses the Rolling guidance filter to achieve multi-scale transformation of the image,and the effectiveness of the proposed algorithm is verified by experiments.3.Based on the multi-scale Rolling guidance filter image fusion,Combine with the sparse expression of images used sparse representation,the image fusion algorithm based on Rolling guidance filter and sparse representation is proposed.Firstly,the dictionary training sample set is decomposed by two layers using a Rolling guidance filter to obtain a low-frequency approximate component sample set and two high-frequency detail component sample sets,and the same two-layer decomposition is performed on the source image.Secondly,the K-SVD algorithm is used to train the two high-frequency detail component sample sets respectively,and the trained dictionary is used to solve the sparse coefficients of the two high-frequency detail components of the source image.Thirdly,the fusion rules are designed for the different scale components of the source image.The low-frequency approximate components adopt the weighted average fusion rule based on image saliency.The high-frequency detail components are enlarged by the L1 norm for the sparse coefficients and reconstructed with the trained dictionary.A fused image of high frequency detail components is obtained.Finally,the fused image is obtained by multi-scale inverse transform of the fused low-frequency approximate component image and the two high-frequency detail component images.Through experiments,the subjective visual effects and objective evaluation indicators are analyzed.The proposed algorithm has better fusion effect.