Fusion of Visible and Infrared Aerial Images Using Deep Learning
Author Information
Author(s): Vipparla Chandrakanth, Krock Timothy, Nouduri Koundinya, Fraser Joshua, AliAkbarpour Hadi, Sagan Vasit, Cheng Jing-Ru C., Kannappan Palaniappan
Primary Institution: University of Missouri, Columbia, MO, USA
Hypothesis
Can a novel end-to-end pipeline effectively register and fuse uncalibrated visible and infrared images?
Conclusion
The proposed DeepFusion pipeline successfully registers and fuses visible and infrared images, improving scene understanding in challenging conditions.
Supporting Evidence
- DeepFusion improves image registration and fusion performance compared to classical methods.
- The proposed wavelet spectral decomposition method effectively extracts relevant features for image matching.
- Keypoint-based analysis shows that DeepFusion retains more original information compared to existing methods.
- Experiments demonstrate the effectiveness of the pipeline across various datasets and conditions.
Takeaway
This study created a new way to combine pictures taken in visible light and infrared, helping us see better in tough conditions like bad weather.
Methodology
The study developed an end-to-end pipeline called DeepFusion that uses wavelet decomposition for image registration and a deep neural network for image fusion.
Limitations
The method relies on the quality of input images and may not generalize well to all scenarios due to variations in sensor characteristics.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website