Authors:
Eric Wisotzky
1
;
2
;
Lara Wallburg
1
;
Anna Hilsmann
1
;
Peter Eisert
1
;
2
;
Thomas Wittenberg
3
;
4
and
Stephan Göb
3
;
4
Affiliations:
1
Computer Vision & Graphics, Fraunhofer HHI, Einsteinufer 37, 10587 Berlin, Germany
;
2
Department of Informatics, Humboldt University, Berlin, Germany
;
3
Fraunhofer IIS, Erlangen, Germany
;
4
Chair of Visual Computing, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany
Keyword(s):
Sensor Array and Multichannel Signal Processing, Deep Learning, Biomedical Imaging Techniques, Image Analysis, Image Upsamling.
Abstract:
Neural network architectures for image demosaicing have been become more and more complex. This results in long training periods of such deep networks and the size of the networks is huge. These two factors prevent practical implementation and usage of the networks in real-time platforms, which generally only have limited resources. This study investigates the effectiveness of neural network architectures in hyperspectral image demosaicing. We introduce a range of network models and modifications, and compare them with classical interpolation methods and existing reference network approaches. The aim is to identify robust and efficient performing network architectures. Our evaluation is conducted on two datasets, ”SimpleData” and ”SimReal-Data,” representing different degrees of realism in multispectral filter array (MSFA) data. The results indicate that our networks outperform or match reference models in both datasets demonstrating exceptional performance. Notably, our approach foc
uses on achieving correct spectral reconstruction rather than just visual appeal, and this emphasis is supported by quantitative and qualitative assessments. Furthermore, our findings suggest that efficient demosaicing solutions, which require fewer parameters, are essential for practical applications. This research contributes valuable insights into hyperspectral imaging and its potential applications in various fields, including medical imaging.
(More)