Our hyperspectral dataset contains 1681 hyperspectral images of different synthetic and historical documents that have all been captured in two spectral ranges, VNIR (Visible and Near-Infrared, 400-1000 nm) and SWIR (Short-Wavelength Infrared, 900-1700 nm).

This dataset has been built with the objective of having a large repository of spectral information of inks, pigments and substrates used historically. Most of the dataset is composed of synthetic samples crafted in recent years with historical recipes, and the rest of the samples are captures of documents preserved in historical archives in Granada, Spain, dating from the 15th to the 20th centuries.

All the samples are MATLAB and Python compatible .h5 files containing the spectral hypercube of the image and a set of metadata. In addition, they are accompanied by a false RGB image and most of them also include their corresponding Ground Truth in order to facilitate segmentation work and the classification and identification of inks and pigments.

Detailed information about the different sets of documents captured for our dataset can be found in this PDF file:

Further details about each of the 1681 samples extracted from them can be found by downloading this EXCEL file:

The full database is stored in a Google Drive folder, accessible upon completion of the following form. If you do not have Python installed, please download the dataset files and then the Visualization Tool we developed. Then, use the tool to visualize the files.

The dataset includes more than 100 different materials. You can see the list and correspondence with the Ground Truth (GT) índices in this EXCEL file:

You can check the filename equivalence of spatially registered VNIR and SWIR minicubes in the database in this Google Docs file:

A visualization tool is also provided to allow observation of the different bands for each sample and the spectra of the known materials. The tool is an .EXE file that you can download here.

Please read these instructions on how to effectively use the visualization tool, which are provided in this PDF file

A tutorial on how to use this tool can be found at:

In case of making any publication in which our dataset is used, please cite it using the reference provided for the following paper:

Hyperspectral dataset of historical documents and mock-ups from 400 to 1700 nm (HYPERDOC), A.B. López-Baldomero, J.L. Nieves, F. Moronta-Montero, M.A. Martínez-Domingo, R. Fernández-Gualda, J. Hernández-Andrés, A.S. Reichert, A. López-Montes, T. Espejo, J. Romero and E.M. Valero, Scientific Data, 12:1248 (2025) https://doi.org/10.1038/s41597-025-05599-0

If you come across any errors, issues, or unexpected behavior while using the dataset or the visualization tool, please let us know by reporting them through this form

For any use of our dataset for commercial purposes, please contact us via email at colorimg@ugr.es