Submit Manuscript  

Article Details


Image Fusion Method Based on Multi-scale Directional Fast Guided Filter and Convolutional Sparse Representation

[ Vol. 12 , Issue. 3 ]

Author(s):

Liu Xian-Hong and Chen Zhi-Bin*   Pages 263 - 269 ( 7 )

Abstract:


Background: A multi-scale multidirectional image fusion method is proposed, which introduces the Nonsubsampled Directional Filter Bank (NSDFB) into the multi-scale edge-preserving decomposition based on the fast guided filter.

Methods: The proposed method has the advantages of preserving edges and extracting directional information simultaneously. In order to get better-fused sub-bands coefficients, a Convolutional Sparse Representation (CSR) based approximation sub-bands fusion rule is introduced and a Pulse Coupled Neural Network (PCNN) based detail sub-bands fusion strategy with New Sum of Modified Laplacian (NSML) to be the external input is also presented simultaneously.

Results: Experimental results have demonstrated the superiority of the proposed method over conventional methods in terms of visual effects and objective evaluations.

Conclusion: In this paper, combining fast guided filter and nonsubsampled directional filter bank, a multi-scale directional edge-preserving filter image fusion method is proposed. The proposed method has the features of edge-preserving and extracting directional information.

Keywords:

Image fusion, fast guided filter, convolutional sparse representation, nonsubsampled directional filter bank, pulse coupled neural network, new sum of modified laplacian.

Affiliation:

Department of Electronics and Optics Engineering, Army Engineering University, Shijiazhuang 050003, Department of Electronics and Optics Engineering, Army Engineering University, Shijiazhuang 050003

Graphical Abstract:



Read Full-Text article