CAVE+KAIST+ICVL

Hyperspectral images contain rich spectral information and have important research value in fields such as material identification, color science, digital cultural relic protection, and all-weather environmental perception (such as autonomous driving and intelligent monitoring). Recovering hyperspectral information from conventional RGB images is a cutting-edge topic at the intersection of computational photography and remote sensing technology. This article uses three commonly used publicly available hyperspectral datasets: CAVE, KAIST, and ICVL. CAVE dataset: Literature: Park J, Lee M, Grossberg M, and Nayar S. 2007. Multispectral imaging using multiplexed illumination. // IEEE 11th International Conference on Computer Vision. Rio de Janeiro: IEEE: 1-8. [DOI: 10.1109/ICCV.2007.4409090] Link: https://cave.cs.columbia.edu/repository/Multispectral KAIST dataset: Literature: Choi I, Jeon D, Nam G, Gutierrez D, and Kim M. 2017. High-quality hyperspectral reconstruction using a spectral prior. ACM Transactions on Graphics, 36(6), 1-13. [DOI: 10.1145/ 3130800.313081] Link: https://vclab.kaist.ac.kr/siggraphasia2017p1/kaistdataset.html ICVL dataset: Literature: Arad B, and Ben-Shahar O. 2016. Sparse recovery of hyperspectral signal from natural RGB images. // European Conference on Computer Vision. Amsterdam: Springer, Cham: 19-34. [DOI: 10.1007/978-3-319-46478-7_2] Link: https://icvl.cs.bgu.ac.il/hyperspectral/ Figures 1 and 2 show the results of simulating spectral reconstruction using different reconstruction methods on the KAIST and ICVL datasets. Figure 3 shows the reconstruction results of the measured compression. Figure 1. Reconstructed image of scene S7 in KAIST dataset, (a) RGB image, (b) simulated compression measurement, (c) spectral density curve, (d) four reconstructed bands Figure 2. Reconstructed image of S2 scene in ICVL dataset, (a) RGB image, (b) simulated compression measurement, (c) spectral density curve, (d) four reconstructed bands Figure 3. Reconstructed images of the S3 scene in the real CASSI measurement dataset, including (a) RGB images, (b) real compressed measurements, and (c) four reconstructed bands In addition, the codes of the references mentioned in this article should also be compiled and uploaded synchronously with the dataset as required by the journal. We have published detailed information on DDXNJUST/Computational-Imaging.