Infrared and Visible Image Fusion in a Multilevel Low-Rank Decomposition Framework Based on Guided Filtering and Feature Extraction

Algorithm 1

Framework of the proposed algorithm for image fusion

Input:

The source of image and .

Output:

Fused image .

/ Part 1: multilevel DLatLRR decomposition. /.

1: for each do

2: for each do

3: Run DLatLRR decomposition on to obtain ,

4: end for

5: end for

 / Part 2: fusion of base parts. /

6: for each do

7: Input image is extracted by the 5th layer of VGG-19 network to acquire ;

8: Transform the -norm of into the activity level map by the Equation (7);

9: Calculate the final weight map via Equations (8) and (9);

10: Use the guided filtering to smooth the final weight to obtain the refined weight via Equations (11) and (12).

11: end for

12: Calculate the fused base parts via Equation (13).

 / Fusion of detail content.

13: for each do

14: Apply the dynamic activity level with the maximum value on to obtain the fused vector as Equation (16);

15: Reconstruct the vector to via Equation (18).

16: end for

 / Reconstruction /.

17: Superpose the fused base part and detail content to reconstruct the fused image , as shown in Equation (19).