Research On Image Fusion Algorithms Based On Morphological Component Analysis

Image fusion technology is a process that integrates multiple input images of the same scene or target into one image by utilizing the complementarity of information.This technology can improve the clarity of the scene or target in the image and provide more accurate and reliable image support for subsequent target recognition and classification.At present,the technology has been successfully applied in many fields such as military,agriculture,security and monitoring.In this thesis,multi-focus image and medical image fusion algorithms based on morphological component analysis(MCA)are mainly studied.The details are as follows:Firstly,the selection methods of the iterations and over-complete dictionaries in morphological component analysis algorithm are improved.In MCA,too small value of the iterations will get incorrect separated results while too large of the iterations will lead to computational efficiency.In order to solve this problem,the noise value of the source image is estimated in this thesis by wavelet transform and mean absolute deviation,which is used as the residual threshold in the process of image decomposition to ensure that the iterations can be chosen according to the characteristics of the image itself.Besides,in view of the weak adaptive ability of the fixed dictionaries of each component in MCA,K-singular value decomposition(K-SVD)algorithm is used in this thesis to train dictionaries in order to make better use of the characteristics of cartoon component and texture component.The experimental results show that the fused image obtained by the improved MCA algorithm performs better in both subjective vision and objective evaluation indexes.Secondly,a multi-focus image fusion algorithm based on improved MCA and non-subsampled Shearlet transform(NSST)is proposed.In order to make up for the deficiency of MCA's detail expression ability,NSST is utilized in this thesis to decompose cartoon components to capture the details of each scale and direction.In addition,a texture fusion rule combining spatial frequency and pulse coupled neural network(PCNN)is proposed,and an improved low frequency fusion rule based on the weighted average of Sum-modified-Laplacian(SML)is proposed according to human visual characteristics.The experimental results show that the algorithm can judge the clear area of multi-focus images accurately,and overcome the artifacts and distortion which affect the image quality in the traditional transform domain algorithms.Thirdly,a medical image fusion algorithm based on improved MCA and multi-objective particle swarm optimization(MOPSO)is proposed.The weighting coefficients of cartoon fusion rules and image reconstruction rules are optimized by MOPSO.And combining with IHS(Intensity,Hue,Saturation)color model,the algorithm is applied to color medical image fusion.The experimental results show that the algorithm can achieve better fusion results than the contrast algorithm in gray medical image fusion and color medical image fusion.