Image Fusion Based On Sparse Representation And Learning Tight Framework

By fusing two or more images in different modes into an information-rich image,image fusion can enhance the visual effect of the fused image while retaining the original image information as much as possible,which facilitates the subsequent visual experience and further image processing and information extraction.At present,image fusion technology has been widely used in biomedicine,remote sensing images,optical photography,navigation mapping and other research fields.The effective representation of image feature information and the fusion method of multi-component information are the key factors to determine the contrast and resolution of the fused image.Therefore,the research of image fusion based on image adaptive representation has important theoretical and practical significance.Although the traditional sparse representation theory can achieve effective image representation,it lacks the characteristics of complete image reconstruction,which makes it difficult to achieve satisfactory results in image fusion.Aiming at the problem of error in sparse decomposition image,this paper proposes an improved morphological component analysis fusion image algorithm.At the same time,an image fusion algorithm based on learning sparse frame is further proposed,which achieves high-quality image fusion while retaining the key features of the original image.The main work of this paper is as follows:(1)The image fusion algorithms in spatial domain and transform domain are deeply studied.The fusion method in spatial domain mainly fuses the pixels in the image point by point through linear weighted fusion and PCA fusion rules.The fusion method in transform domain uses some known transformation method to expand the image on a set of bases to realize the multi-band decomposition of the image,and then designs appropriate fusion rules according to the characteristics of each band to realize the fusion of coefficients.The corresponding inverse transform is used to reconstruct the fused image.(2)Because of the sparse representation theory based image fusion algorithm,there will be a certain degree of sparse approximation in the process of coefficient solving,resulting in the fused image blurred phenomenon.This paper proposes an improved image fusion algorithm based on morphological component analysis combined with sparse representation theory.The fused image is taken as the original training data,and an adaptive dictionary is obtained by learning strategy.The original image is approximately decomposed into texture image and cartoon image by morphological component analysis.According to the image fusion rule of cartoon component design ?1 norm,the fusion rule of absolute value of texture component is designed.To avoid introducing noise and error into the decomposition process,the residual components after decomposition are fused.Finally,the final fused image is obtained by superimposing these components.Experiments show that this method can avoid the errors introduced by the sparse process,and realize image fusion by using the features of different image morphological components,which preserves the details of the original image,(3)Aiming at the problem of low efficiency of dictionary training and lack of complete signal reconstruction in traditional sparse representation,this paper proposes an image fusion algorithm based on training tight framework,and designs a new fusion rule for tight framework.Firstly,the data blocks are extracted from the fused image,and the samples needed to train the adaptive tight frame are constructed,and the tight frame is obtained by learning strategies under orthogonal constraints.Then,the tight frame is used to realize the multi-band representation of the image.According to the activity of the tight frame filter,it is divided into two categories,i.e.low-pass filter combined with high-pass filter banks,and the output of these two groups of filters is utilized.Secondly,low-frequency image and high-frequency image are constructed.Secondly,low-frequency image is fused with ?1 norm-based fusion rules.For each pixel value of high-frequency image,a fusion rule based on gradient summation and comparison is designed to realize high-frequency component fusion.Finally,the fused image is obtained by using the fused low-frequency component and high-frequency component.Experiments show that this method not only preserves image edge information,but also improves the efficiency and accuracy of fusion.