Research On The Image Fusion Algorithm Based On Non-sumsampled Shearlet Transform

Image fusion is an emerging research hotspot which involves multiple fieldssuch as: information fusion, sensors, image processing, etc. In the field of fusionresearch, how to efficiently represent and analyze image is one of a core issue.Effectiveness of image representation method directly determines the quality of thefused image.With flexible direction features and a multi-resolution structure,Non-subsampled Shearlet Transform is currently the most advanced method.Compared with other multi-scale transformation method, NSST has more obviousadvantages. From the perspective of approximation theory, NSST is an optimalapproach, which is a true representation of sparse image at all directions and alldifferent scales.In addition, different from traditional shearlet transform, NSST has a translationinvariance. Without down sampling operation, the pseudo-Gibbs effect can beovercomed when the image is reconstructed.In this paper, the theory of NSST is deeply studyed and its application in thefield of image fusion is discussed. The main research and innovation of this paper isas follows:(1) Image Fusion Algorithm of Mixed Multi-scale Analysis Based on NSSTSingle multi-scale decomposition methods are just good at dealing with a certaintype of feature of an image, but not suitable to other types of the problem. In thispaper, the complementary characteristics of different multi-scale analysis methods arestudied, and two kinds of image fusion algorithm of mixed multi-scale analysis basedon NSST are proposed. The algorithm includes the following three steps. Firstly, thesource images are decomposed with NSST; secondly, the low-coefficients aredecomposed again with DWT and SWT which have the complementarycharacteristics; thirdly, the final decomposed coefficients are fused according to thedesigned fusion rules; then the fused coefficients are used to image reconstruction byusing the inverse DWT and SWT; finally, the inverse NSST is used to reconstruct thefused images. The experimental results show that the two algorithms in this paper are superiorto the traditional single multi-scale fusion method from the subjective effect and theobjective parameters. Meanwhile, comparing the two kinds of algorithms, we can findthat image fusion algorithm that combine NSST with SWT can effectively inhibit thepseudo-Gibbs effects due to downsampling operation and has a greater advantage incapturing more feature information of source images.(2) Image Fusion Algorithm Based on NSST and Compressed SensingAfter the image decomposition with NSST, the high-frequency coefficients havea large amount of data and greater sparsity. In order to obtain fusion results rapidly,an image fusion algorithm based on NSST combined with CS is presented. Firstly, thesource images are decomposed with NSST; secondly, the high-frequency sub-bandcoefficients of the decomposed images are compressed, fused and reconstructed byCS; then, based on ocal area variance and local area energy the low-frequencycoefficients was fused; finally, the inverse NSST is used to get the final fused image.Due to only the compressed values of the high frequency coefficients are fused,the image fusion effects can’t be affected, and the running time of the algorithm canbe reduced. In this paper, the multi-focus image, medical image and infrared andvisible images are used to verify the effectiveness of the algorithm. The simulationresults indicate that this algorithm can achieve the fusion of the image without priorknowledge of the original image. When the image size is larger, although the fusionimage quality is sacrificed, it can significantly improve the speed to reduce the timecost and hardware requirements. The algorithm provides an idea on how to satisfy thereal time requirements in the fusion system, and which has a great practical value.(3) Image Fusion Algorithm Based on NSST and NMFAfter the image decomposition with NSST, the transform coefficients have alarger redundancy. In order to reduce the redundant information, an image fusionalgorithm based on NSST and non-negative matrix factorization (NMF) is introduced.Firstly, the NSST is used decompose image; secondly, the low-frequency coefficientsare fused with NMF; then Local Aera Maximum SML was used to fuse thehigh-frequency coefficients; finally, the inverse NSST is used to reconstruct the final fused image.The proposed algorithm can effectively remove redundant information, extractglobal features and capture more direction details information of multi-source image.Experiments show that the proposed algorithm has obvious advantages and the fusedimage quality has been greatly improved. The proposed algorithm is superior to otherfusion algorithms both from the subjective effect and objective parameters.