Sharma, H and Koundinya, S and Sharma, M and Upadhya, A and Manekar, R and Mukhopadhyay, R and Karmakar , A and Chaudhury , S
(2018)
2D-3D CNN based architectures for spectral reconstruction from RGB images.
In: Conference on Computer Vision and Pattern Recognition (CVPR) Workshop ,2018, June 18-22, 2018, Salt Lake City .
(Submitted)
Abstract
Hyperspectral cameras are used 10 preserve fine spectral details of scenes that are not captured by traditional RGB cameras that comprehensively quantize radiance in RGB images. Spectral details provide additional information that improves the performance of numerous image-based ana lytic applications, but due to high hyper. spectral hardware cost and associated physical constraints, hyperspectral im ages are not easily available for.fiirther processing. Moli voted by the performance of deep learning for various com puter vision applications, we propose a 2D convolution neu rail network and a JD convolution neural net/Work-based ap approaches for hyperspec!ral image reconstruction from RGB images. A 2D-CNN model primarily focuses on ex/ racing g spectral data by considering the only spatial correlation of the channels in the image, while in the JD-CNN model the inter channel co-relation is also exploited to refine the extraction of spectral data. Our JD-CNN has an architecture that achieves very good performance in terms of MRAE and RMS£. In contrast to JD-CNN, our 2D-CNN based architecture also achieves comparable performance with very less computa tional complexity.
Actions (login required)
 |
View Item |
Download Statistics
Download Statistics