Convolution dictionary learning for visible-infrared image fusion via local processing

Although convolutional sparse coding overcomes the limitations of block-based sparse representation in the process of convolutional dictionary learning, it relies too much on the Alternating Direction Method of Multipliers(ADMM) formula in the Fourier domain. This leads to a loss of locality, impacting the quality of the fused image which tends to have a fuzzy texture. To solve this problem, we use a local processing convolution dictionary-learning method to obtain a dictionary and apply the obtained dictionary to the fusion of visible-infrared images. The proposed method not only solves the problem of global convolutional sparse coding, but also overcomes the blurry fusion image defect. Experimental results show that the proposed method is superior to the comparison methods in subjective and objective evaluation. Compared with deep learning fusion methods, our fusion framework achieves an average improvement of 5.18%, 5.02%, and 4.77% in the objective evaluation QAB/F, Qe and Qp, respectively. (C) 2021 The Authors. Published by Elsevier B.V.

Automated summary

The proposed local processing convolution dictionary-learning method improves image fusion quality by addressing the limitations of global convolutional sparse coding and overcoming the blurry fusion image defect. Experimental results show superior performance compared to other methods, with an average improvement of 5.18%, 5.02%, and 4.77% in objective evaluation metrics.