Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
June 18th, 2021
Hyperspectral Reflectance Imaging hypercubes include remarkable information into a large amount of data. Therefore, the request for automated protocols to manage and study the datasets is widely justified. The combination of Spectral Angle Mapper, data manipulation, and a user-adjustable analysis method constitutes a key-turn for exploring the experimental results.
This protocol offers new opportunities involving the analysis of paintings of faces. It supports the users step by step along the whole data analysis process. This protocol has two main advantages.
The users can tailor the analysis according to their preferences, and for the first time, spectrum manipulation has been introduced to analyze the hyperspectral data. Hyperspectral reflectance imaging is successfully used to study skin disease or tumor diagnosis. Even if a protocol was born in the counter heritage field it can be applied to clinical health dataset as well.
With the support of art experts, perform a preliminary inspection of the painted surface to identify the main features of the painting. Make a note of the pictorial techniques used by the artist, the different brush strokes of paint on the canvas, and qualitatively estimate the characteristics of the brush strokes with particular attention to their size. Create ad hoc samples where the brush strokes show characteristics similar to those applied by the artist by mimicking the pictorial technique used by the artist.
Acquire the hyperspectral data and check whether the spatial resolution of the hypercubes can distinguish the different brushstrokes on the RGB images of the painted surface. Run PointSel, the isolated measuring point selection code to manually select some reference spectra on the surfaces of the test samples. Type the command line including the semicolon in the terminal window and press enter to run the code.
Select the measurement points by clicking the interactive window that one by one shows the two dimensional RGB images of the fields of view. Run SAM_Standard, the standard SAM maps evaluation code to extract the SAM maps using the whole spectra. Type the command line, including semicolon in the terminal window, then press enter to run the code.
SAM maps are saved as PNG images in the current work folder. Check whether the obtain similarity maps display the details of the brush strokes used to realize the test samples. If not, restart the process by readjusting the distance between the test sample surface and the acquisition equipment.
According to the evaluation obtained by the test samples set the distance between the surface of the painting under investigation and the acquisition equipment. Perform IO of the hyperspectral data by organizing, reading and managing the hypercubes. Run the HS FileLister code to store the list of the files containing the hypercubes and the related information into two variables at the disposal of the algorithm.
Run the HS_Crop PNG code to select the portion of each FOV to be used in the analysis of the data. Next, run the PointSel code and click within the displayed interactive window to identify the reference spectra as isolated measuring points over the surface of the monitored areas. Type the command line including the semicolon in the terminal window and press enter to run ReticularSel, the Reticular selection code.
This automatically selects the reference spectra as a regular reticulum of measuring points superimposed to the surface of the monitored areas. This selection method makes the analysis very time consuming as the number of references is large. Type the command line including the semicolon in the terminal window and press enter to run SaveImPoint.
This saves a location at the selected measurement point superimposed to the pictures of the fields of view. Run Spectra_Importer, the external references importer code to create a variable containing references from data sets and databases independent from the hypercubes acquired on quarto stato. Note that the spectra have different size with respect to those obtained with the hyperspectral camera.
Run the SAM complete code to evaluate the similarity maps. Feed the code with the desired pre-processing option entering zero or one in the dialogue box. Zero to require the spectra normalization only or one to require that after the normalization the spectra are derived one time.
enter this sequence of numbers corresponding to the desired columns of the references matrix into the dialogue box by typing the numbers separated by a white space. Press Enter to continue. Set the method to zero for no manipulation of the data.
One to require manual selection of the wavelength ranges of the spectra to be considered before starting the analysis or two to require the algorithm to order the data based on a specific criterion and before the evaluation of the SAM maps. To select the end members for the SAM analysis, the algorithm either retrieves the references spectra among the hypercubes by manually selecting some isolated measuring points or automatically samples the painting surface, providing a particular selection of measuring points within one or more FOVs. The algorithm can also compare the hypercubes with external spectra like the ones obtained by a portable FORS miniature spectrometer.
When the pre-processed references appear on an interactive window, one or more wavelength intervals to be analyzed can be manually selected. In automatic selection, the algorithm computes a maximum variance within the desired references and orders spectra according to this criterion. If the maximum variance corresponds to the nth wavelength the content of the nth component of each pre-processed spectrum will be moved to the first position of a rearranged hypervector and so on.
Following the automatic manipulation, the algorithm applies a floating threshold to the variance values and evaluates the SAM maps at the increasing threshold which results in a total of two n plus one sets of maps where n is a number of values assumed by the threshold. The obtained similarity maps provide new insights into the details of the mapped area. They can help to compare the samples and the references.
The possibility to customize the analysis and to exploit any spectrum as reference, expands the user horizons but simultaneously asks the user for a careful evaluation of their choices This approach allows the use of spectrum manipulation as an analysis tool, therefore computer vision and statistical studies can help deepen the knowledge about the possibility of this matter.
View the full transcript and gain access to thousands of scientific videos
Overview
This protocol presents a systematic approach for analyzing hyperspectral reflectance imaging (HRI) data from paintings, with a focus on the iconic artwork Quarto Stato by Giuseppe Pellizza da Volpedo. The method leverages Spectral Angle Mapper (SAM) algorithms and introduces spectrum manipulation to efficiently process and interpret large HRI datasets, enabling detailed mapping of painting features such as brushstrokes and pigment distribution.
Key Study Components
Area of Science
- Art conservation science
- Imaging spectroscopy
- Data analysis in cultural heritage
Background
- Reflectance Spectroscopy (RS) and Fiber Optics Reflectance Spectroscopy (FORS) are established non-invasive techniques for analyzing artworks.
- Hyperspectral Reflectance Imaging (HRI) produces large 3D datasets (hypercubes) associating spectral and spatial information from painted surfaces.
- Manual analysis of these datasets is time-consuming and requires expert knowledge, especially for large-scale studies.
- Automated and robust data analysis protocols are needed to fully exploit the information contained in HRI datasets.
Purpose of Study
- To develop and demonstrate a protocol for efficient, customizable analysis of HRI data from paintings.
- To introduce spectrum manipulation as a novel tool for hyperspectral data analysis in art conservation.
- To enable both manual and automated selection of reference spectra for detailed mapping of painting features.
Methods Used
- Preliminary inspection of the painting surface with art experts to identify key features and brushstroke characteristics.
- Acquisition of hyperspectral data and assessment of spatial resolution to distinguish brushstrokes.
- Manual and automated selection of reference spectra using custom codes (PointSel, ReticularSel).
- Application of Spectral Angle Mapper (SAM) algorithms to generate similarity maps between sample and reference spectra.
- Implementation of spectrum manipulation, including normalization, derivation, and wavelength range selection, to enhance analysis.
- Comparison of hypercube data with external reference spectra from databases or portable FORS devices.
Main Results
- The protocol enables detailed mapping of brushstroke patterns and pigment distribution in paintings.
- SAM-based similarity maps reveal subtle features and differences across the painted surface.
- Spectrum manipulation enhances the discrimination of spectral features and supports customized analysis.
- The approach is adaptable for both manual and automated workflows, accommodating user preferences and dataset size.
- The protocol is applicable beyond art conservation, with potential uses in clinical imaging and other fields.
Conclusions
- The protocol provides a robust, flexible framework for analyzing large HRI datasets from artworks.
- Spectrum manipulation and automated reference selection expand analytical capabilities and user control.
- This methodology supports deeper insights into painting techniques and materials, aiding conservation and research efforts.
Frequently Asked Questions
Executive Industry Relevance
Automated hyperspectral reflectance imaging (HRI) analysis protocols, originally developed for art conservation, offer scalable solutions for high-dimensional data interrogation in biopharma discovery. The integration of spectrum manipulation and robust similarity mapping enables precise, reproducible quantification of spatially resolved molecular features, supporting early-stage target validation and mechanistic de-risking. These capabilities are directly relevant for workflows requiring high-throughput, quantitative, and spatially explicit analysis of complex biological samples.
Strategic Applications in Biopharma R&D
Early Discovery & Target Validation
- Enables systematic interrogation of spatially heterogeneous biological samples for target validation.
- Supports mechanistic de-risking by mapping molecular distributions with high spatial fidelity.
- Facilitates predictive confidence through quantitative, reproducible spectral similarity analysis.
Screening & Assay Development
- Prepares validated, spatially resolved datasets for downstream screening workflows.
- Standardizes assay outputs by leveraging automated spectrum selection and manipulation protocols.
- Enables reproducible, quantitative readouts for compound evaluation in complex matrices.
Translational & Preclinical Research
- Aligns spatially resolved molecular mapping with translational biomarker discovery when supported by sample type.
- Provides continuity from discovery through preclinical validation by supporting robust, quantitative imaging analytics.
- Reduces risk in advancement decisions by enabling detailed, quantitative mapping of biological features.
Pipeline & Workflow Integration
HRI-based protocols fit within the discovery-to-preclinical continuum, supporting workflows from early hypothesis testing to advanced analytics for lead identification and translational research.
- Discovery Biology: Facilitates hypothesis testing and pathway clarification via spatially resolved spectral analysis.
- Screening: Delivers assay-ready, reproducible, and quantitative outputs for high-throughput workflows.
- Analytics: Provides robust similarity mapping and statistical outputs for comparative condition analysis.
- Translational Research: Supports biomarker alignment and preclinical continuity when applied to disease-relevant systems.
- Enterprise Reuse: Establishes a scalable, adaptable imaging analytics capability for diverse R&D applications.
Operational & Enterprise Impact
- Scientific Value: Increases predictive confidence and reduces mechanistic ambiguity in spatially complex samples.
- Operational Value: Standardizes and automates high-dimensional data analysis for reproducibility and scalability.
- Strategic Value: Improves go/no-go decision quality and capital efficiency by enabling robust, quantitative analytics.
- Portfolio Impact: Supports risk-adjusted prioritization and advancement across discovery and preclinical pipelines.
Implementation Considerations
- Requires expertise in hyperspectral imaging and spectral data analysis.
- Demands access to high-resolution imaging instrumentation and computational infrastructure.
- Necessitates cross-team standardization of data acquisition and analysis protocols.
- May require adaptation for different biological model systems or sample types.
- Analysis throughput and interpretability depend on robust reference spectra and algorithmic parameterization.
