TY - GEN
T1 - Accurate spectral super-resolution from single RGB image using multi-scale CNN
AU - Yan, Yiqi
AU - Zhang, Lei
AU - Li, Jun
AU - Wei, Wei
AU - Zhang, Yanning
N1 - Publisher Copyright:
© Springer Nature Switzerland AG 2018.
PY - 2018
Y1 - 2018
N2 - Different from traditional hyperspectral super-resolution approaches that focus on improving the spatial resolution, spectral super-resolution aims at producing a high-resolution hyperspectral image from the RGB observation with super-resolution in spectral domain. However, it is challenging to accurately reconstruct a high-dimensional continuous spectrum from three discrete intensity values at each pixel, since too much information is lost during the procedure where the latent hyperspectral image is downsampled (e.g., with 10 scaling factor) in spectral domain to produce an RGB observation. To address this problem, we present a multi-scale deep convolutional neural network (CNN) to explicitly map the input RGB image into a hyperspectral image. Through symmetrically downsampling and upsampling the intermediate feature maps in a cascading paradigm, the local and non-local image information can be jointly encoded for spectral representation, ultimately improving the spectral reconstruction accuracy. Extensive experiments on a large hyperspectral dataset demonstrate the effectiveness of the proposed method.
AB - Different from traditional hyperspectral super-resolution approaches that focus on improving the spatial resolution, spectral super-resolution aims at producing a high-resolution hyperspectral image from the RGB observation with super-resolution in spectral domain. However, it is challenging to accurately reconstruct a high-dimensional continuous spectrum from three discrete intensity values at each pixel, since too much information is lost during the procedure where the latent hyperspectral image is downsampled (e.g., with 10 scaling factor) in spectral domain to produce an RGB observation. To address this problem, we present a multi-scale deep convolutional neural network (CNN) to explicitly map the input RGB image into a hyperspectral image. Through symmetrically downsampling and upsampling the intermediate feature maps in a cascading paradigm, the local and non-local image information can be jointly encoded for spectral representation, ultimately improving the spectral reconstruction accuracy. Extensive experiments on a large hyperspectral dataset demonstrate the effectiveness of the proposed method.
KW - Convolutional neural networks
KW - Hyperspectral imaging
KW - Multi-scale analysis
KW - Spectral super-resolution
UR - https://www.scopus.com/pages/publications/85057186401
U2 - 10.1007/978-3-030-03335-4_18
DO - 10.1007/978-3-030-03335-4_18
M3 - 会议稿件
AN - SCOPUS:85057186401
SN - 9783030033347
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 206
EP - 217
BT - Pattern Recognition and Computer Vision - First Chinese Conference, PRCV 2018, Proceedings
A2 - Liu, Cheng-Lin
A2 - Tan, Tieniu
A2 - Zhou, Jie
A2 - Lai, Jian-Huang
A2 - Chen, Xilin
A2 - Zheng, Nanning
A2 - Zha, Hongbin
PB - Springer Verlag
T2 - 1st Chinese Conference on Pattern Recognition and Computer Vision, PRCV 2018
Y2 - 23 November 2018 through 26 November 2018
ER -