| Due to the singularity of image information obtained by different sensors,it is difficult to describe the object accurately.In this context,image fusion technology has been widely concerned by researchers.The technique synthesizes the image information taken from the same scene under different sensors to generate a description of the scene,which cannot be obtained from a single source image information.It can effectively synthesize the complementarities of image information obtained by different sensors and provide a more accurate description for the object of observation.Therefore,the technology has been successfully applied to many fields,such as medical imaging,remote sensing,military defense,machine vision and so on.In the existing image fusion methods,a single dictionary is usually used to represent the different components of the image.Because different components of images have different characteristics,it is difficult for a single dictionary to effectively represent different components.Most of the studies assume that the source image is obtained in the case of accurate registration and no noise,which obviously can not meet the needs of reality.In this paper,three image fusion algorithms based on multi-component analysis are proposed to overcome the shortcomings and shortcomings of the existing image fusion methods.（1）For the fusion of noise-free images,a multi-source image fusion algorithm based on multi-component dictionary learning and cartoon texture decomposition is proposed.In this method,the image decomposition paoblem is transformed into the image classification problem,and a cartoon-texture discriminant dictionary learning model is designed.Considering the fact that the image decomposition is not only related to the dictionary studied,but also related to the decomposition strategy.A new image decomposition model is designed,in which texture components are regarded as noise superimposed on cartoon components of source images,and consistent canonical terms of similarity of non-local mean are introduced.Finally,the dilution coding coefficients of fused images are selected according to the maximum sparse coding sparse l1 norm of the corresponding components.（2）For the fusion of the noise image,propose a discriminative dictionary learning based on fusion and the details of image noise protection component decomposition method.In order to overcome the shortcomings of traditional dictionary learning method,in the design of dictionary learning model,the image is modeled as a superposition of cartoon composition and texture component.By introducing the weighted nuclear norm regularization term to describe texture ingredients,and add constraints on the coefficient of cartoon components,to improve the discrimination ability of different components of the dictionary.In order to put the image components separation in image decomposition model,we introduce a new sparse kernel weighted Schatten norm regularization method to extract texture components.Self similarity estimation using the non noise texture component the local image,while using the gradient Affirmative map based on constraints to maintain retention in cartoon noise.Finally,fuse the coefficients of different components（3）The fusion rules of the first two methods are relatively simple.Therefore,an image fusion method based on the combination of low rank image and sparse component decomposition and Pulse coupled neural networks（PCNN）is proposed for noise-free images.Based on the low rank decomposition and the sparse representation,the sparse coding coefficients of the corresponding components are obtained by sparse representation of different components in different dictionaries.In the fusion process,in order to preserve the brightness information of the source image,a"absolute value"strategy is used to fuse the low-rank components.The characteristic gradient is used to excite PCNN and the coefficients of large ignition times are selected as the sparse component coding coefficients of fused images. |
