Ink identification using only spectral reflectance information poses significant challenges due to material degradation, aging, and spectral overlap between ink classes. This study explores the use of hyperspectral imaging and machine learning techniques to classify three distinct types of inks: pure metallo-gallate, carbon-containing, and non-carbon-containing inks. Six supervised classification models, including five traditional algorithms (Support Vector Machines, K-Nearest Neighbors, Linear Discriminant Analysis, Random Forest, and Partial Least Squares Discriminant Analysis) and one Deep Learning-based model, were evaluated. The methodology integrates data fusion from different imaging systems, sample extraction, ground truth creation, and a post-processing step to increase uniformity. The evaluation was performed using both mock-up samples and historical documents, achieving micro-averaged accuracy above 90% for all models. The best performance was obtained using the DL-based model (98% F1-score), followed by the Support Vector Machine model. In the case study documents, the overall performance of the traditional model was better. This study highlights the potential of hyperspectral imaging combined with machine learning for non-invasive ink identification and mapping, even under challenging conditions, contributing to the conservation and analysis of historical manuscripts.

HYPERDOC is a hyperspectral imaging dataset of historical documents and mock-ups, designed to facilitate research in material identification in the cultural heritage domain. It contains mock-ups of historical inks (metallo-gallate, sepia, carbon-based, and mixtures) on various supports, including some artificially aged, and historical documents from the 15th to 17th centuries (manuscripts, illuminated manuscripts, and family trees). Hyperspectral reflectance images were acquired using line-scan cameras in the VNIR (400-1000 nm) and SWIR (900-1700 nm) ranges and were spatially registered. Small regions of interest, referred to as ‘minicubes’, were extracted from the full document images, and pixel-level ground truth material annotations were performed. False-color RGB images and metadata were included in both the full document and minicube captures. The HYPERDOC dataset has been successfully applied in various experimental studies, including ink classification using machine learning models, spectral unmixing, colorimetric analysis, and binarization. These applications highlight the dataset’s potential, which is publicly available to promote interdisciplinary collaboration and advance the use of hyperspectral imaging in the conservation field.

Characterizing pigments and dyes in historical manuscripts is challenging due to the fragility of materials, the complex composition of low-concentration elements, and sampling limitations. Consequently, complementary non-invasive analytical techniques and non-contact measurement methods are often required. This study presents the most comprehensive spectral database to date, combining diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) and hyperspectral imaging (HSI) to aid in identifying pigments, dyes, and binders historically used in illuminated and decorated manuscripts. A total of 156 painting mock-ups were created using traditional techniques, incorporating variations in binders, pigment particle sizes, support types, surface roughness, and application methods. Spectral imaging was performed in the visible and near infrared (VNIR) and short-wave infrared (SWIR) regions, while DRIFTS analysis covered the middle wave infrared (MWIR) region. For DRIFTS, both contact and non-contact measurements were tested. Using the samples in the database, the influence of binder, support, and grain size on the sample spectra and color were analyzed and discussed. This database facilitates pigment and dye identification using DRIFTS or HSI data independently or in combination through data fusion, applying techniques ranging from direct spectral comparison to advanced methods such as machine learning and spectral unmixing. By making this database publicly available, the study underscores the value of DRIFTS and HSI in identifying painting materials and contributes to the preservation of historical manuscripts.

The aim of this work is to provide the cultural heritage community with a comprehensive hyperspectral image database of handwritten laboratory samples, including various writing inks commonly found in historical documents. The database contains 195 samples registered in the VNIR (400-1000 nm) and SWIR (900-1700 nm) spectral ranges, along with complete information about the ink recipes (components and concentrations used for each ink and mixture), and their corresponding Ground Truth images. The database is now publicly available as part of a bigger database related to the Hyperdoc project and can be used to perform different tasks. We present here one example: the classification of iron gall vs non-iron gall inks.

Temperature distribution can be acquired through non-contact temperature measurement using multispectral imaging. However, the challenge lies in radiometric temperature inversion owing to the unknown emissivity. Despite the promising results demonstrated by traditional algorithms and neural networks, enhancing the precision and reliability of temperature inversion remains a challenge. To tackle these challenges, in this work, we propose the use of ensemble learning for temperature distribution inversion in infrared multispectral imaging. The network comprises a base-learner and a meta-learner, trained to establish the nonlinear relationship between temperature and multispectral distribution measurements. Moreover, the network architecture exhibits high robustness against noise arising in the testing environment. Simulations and real experiments on multispectral imaging measurements illustrate that ensemble learning can be a potent tool for multispectral imaging radiation temperature distribution measurement, achieving superior inversion performance compared to other neural networks. The reproducible code will be available at https://github.com/shuowenyang/Temperature-Inversion.