Infrared & visible images fusion based on redundant directional lifting-based wavelet and saliency detection

• Published in: Infrared physics & technology 2019-09, Vol.101, p.45-55
• Main Authors: Minghui, Song, Lu, Liu, Yuanxi, Peng, Tian, Jiang, Jun, Li
• Format: Article
• Language: English
• Subjects: Image fusion; Redundant directional lifting-based wavelet; Saliency detection; Weighted map
• ISSN: 1350-4495; 1879-0275
• DOI: 10.1016/j.infrared.2019.05.017

•Novel image fusion frameworkfor infrared and visible images.•Novel nonsubsampled redundant directional lifting-based wavelet transform.•Novel hybrid saliency detection algorithm which incorporateslocal contrast and global contrast information.•Guided image filter is utilized to obtain spatial consistency to avoid block artificialities.•State-of-the-art performance achieved in terms of both subjective and objective evaluations. This paper proposes a novel infrared and visible image fusion algorithm, based on redundant directional lifting-based wavelet (RDL-Wavelet) and a hybrid saliency detection algorithm. Firstly, the input images are decomposed into approximation coefficients matrix and detail coefficient matrices. Comparing to traditional lifting-based wavelet, the proposed method involves no subsample processes, hence Gibbs artificialities can be alleviated with its shift-invariant property. Furthermore, it can adapt far better to the image orientation features in which predicting and updating signals can be derived even at the fractional pixel precision level, with the help of ‘Sinc’ interpolation technique. Secondly, a fusion rule based on saliency detection, which concerning both local contrast and global contrast informations, is adopted to obtain the result’s approximation and detail coefficients matrices. In local saliency detection process, morphology close operation is used to generate the refined saliency map. Finally, inverse RDL-Wavelet serves to regain fused image. Meanwhile, a feasible optimization scheme can be used to further enhance the computing efficiency of proposed algorithm. Experiments show that our method gives improved quantitative and qualitative results compared with recently developed methods.