Research on mixed scale image fusion algorithm for dual modal night images

DOI:	10.11809/bqzbgcxb2024.05.040
Keywords:	image fusion; mixed scale; convolutional sparse representation; infrared image; visible image
Abstract:	Aiming at the problems of detail blur, low contrast and loss background information in the traditional infrared and visible image fusion algorithm, this paper proposes a new infrared and visible image fusion algorithm based on mixed scale. Firstly, the source images are decomposed by latent low rank representation decomposition to obtain the low rank sub band saliency sub band, respectively. Secondly, through non subsampled contourlet transform, the infrared and visible image low rank part are decomposed into low frequency and high frequency sub bands, separately. Thirdly, the saliency sub bands are fused by using convolutional sparse representation rules. Then, combining the global and reginal mean value, region energy to fuse the low frequency sub band. Finally, the weighted decision map diagram is adopted to merge the low frequency sub band. Experimental of self built datasets and public datasets results show that the proposed algorithm can fully reserve the source image effective information, with more image balance contrast and sharpness, richness of detail information, superior to other five efficient image fusion algorithms in terms of subjective and objective evaluations.
Published:	2024-05-31
PDF