Technique for image fusion based on non-subsampled contourlet transform domain improved NMF

Abstract

To overcome the problem of multi-sensor image fusion, a technique for image fusion based on non-subsampled contourlet transform (NSCT) domain improved nonnegative matrix factorization (NMF) is presented. Firstly, by using NSCT, multi-scale and multi-direction sparse decompositions of source images are performed. Then, an improved NMF technique is utilized to complete the fusion of low-frequency sub-images. The low-frequency fused image can be produced fast by the process which does not involve the randomization of the vectors W and H at all, in addition, the fusion course of high-frequency sub-images can be dealt with by use of the model of adaptive unit-fast-linking pulse coupled neural network (AUFLPCNN). Finally, the ultimate fused image can be obtained by synthesizing all sub-images with inverse NSCT. The simulated experiments show that the technique is effective.Keywords image fusion, non-subsampled contourlet transform, nonnegative matrix factorization, singular value decomposition CitationKong W W, Lei Y J, Lei Y, et al. Technique for image fusion based on non-subsampled contourlet transform domain improved NMF.

Pan-sharpening: a fast variational fusion approach

“…We compare our results with those of the state-of-the-art fusion approaches based on ATWT and NSCT because of their high fusion performances [7][8][9][10]. (a) and (b) of Figures 5-7 are the input images, (c)-(e) of Figures 5-7 show the results obtained by ATWT, NSCT and our model.…”

“…Multi-resolution analysis (MRA) [1] is one of the efficient tools which decomposes the pan and the multi-spectral image in different scale spaces and designs suitable rules for fusion, but Gibbs effects and canvas-like pattern usually appear near the edges of the fused image. This impairment is effectively addressed byà trous wavelet transform (ATWT) [2], the curvelet transform [3,4] and the non-subsampled contourlet transform (NSCT) [5][6][7][8][9][10]. Recent studies show that both ATWT and NSCT turn out to be more suitable and more effective for fusion because of their shift-invariant [7][8][9][10].…”

“…In [9], the super-resolution technique is combined with NSCT for Pan-sharpening while efficiently preserving the texture and contour information of the PAN. In [10], NSCT is combined with the improved nonnegative matrix factorization for better fusion effects.…”

Technique for image fusion based on NSST domain improved fast non-classical RF

“…In addition, six current methods are utilized to compare with M7 in this paper, which are Shearlet-based method (M1) [23], NSST-based method (M2) [26], LC-PCNN-based method (M3) [6], RF-based method (M4) [14], NSCT-RF-based method (M5) [27] and NSCT-NMF-based method (M6) [28]. In order to make the comparison persuasive and reliable, the parameters settings in M1-M6 are still implemented according to the content in Ref.…”

“…In order to make the comparison persuasive and reliable, the parameters settings in M1-M6 are still implemented according to the content in Ref. [23,26,6,14,27,28]. If you want to obtain more details, please refer to related references mentioned above.…”