Infrared-Visible Image Fusion based on Stacked Sparse Autoencoder and Non-Subsampled Contourlet Transform

Fusing the infrared and visible images is still an important issue within images fusion. This paper develops a novel algorithm which combines the Stacked Sparse Autoencoder (SSAE) with Non-Subsampled Contourlet Transform (NSCT) to present the fusion of infrared and visible images for selecting efficient information. The SSAE is used to learn the high-level features from the images which are going to be fused, and these features are clustered by Expectation-Maximization(EM). After clustering, these features which are going to be deal with a decision rule produce a decision map. The NSCT is used to process infrared and visible images, and obtain approximation and detail information. Finally, acquiring fusion image based on combining approximation information with detail information which either is selected or added by guiding with decision map. The experimental data reveal that algorithm presents some better performances than general methods with objective and subjective evaluation standard respectively.