Suspectral

Visualization and reconstruction of hyperspectral imaging data

from RGB images based on spectral response functions.

Features

• Support for hyperspectral images in the ENVI format. 📂
• Sensor-based synthesis of RGB images from hypercubes. 🖼️
• Preview of sRGB images based on the CIE XYZ 1931 CMFs. 👁️
• Options for performing color correction on synthetic images. 🎨
• Ability to display pixel spectra either individually or in groups. 📈
• Export of spectral data into CSV, MATLAB, and NumPy tabular files. 📝
• Saving image previews and spectral plots into PNG, JPG, TIFF, and BMP. 📷
• Transformations for image preview: zooming, panning, rotations, flips. 🔄️
• Automatic detection of the operating system's light and dark themes. 🔥

Running / Building

This repository contains two subprojects: suspectral-app and suspectral-notebook. Both projects are intended to be run separately, each in its own virtual environment, if needed.

suspectral-app

This subproject contains the software application developed using PySide6 bindings for the Qt framework. To run the software, simply follow these steps from the suspectral-app directory, assuming you have already activated its virtual environment:

1. pip install -r requirements.txt
2. pyside6-rcc resources/resources.qrc -o resources.py
3. py app.py

To run automated tests using pytest, simply execute one of the following commands:

• pytest tests (without coverage)
• pytest tests --cov=suspectral (with coverage)

The software can be packaged into a Windows executable using PyInstaller:

1. pyinstaller app.spec

To create an installation wizard via Inno Setup, use the provided compilation script.

suspectral-notebook

This subproject contains the Jupyter notebooks contain various experiments and demonstrations revolving around image synthesis and spectral reconstruction. To run the notebooks, simply follow these steps from the suspectral-notebook directory, assuming you have already activated its virtual environment:

1. pip install -r requirements.txt
2. jupyter notebook

You must download datasets from their respective authors and extract them in corresponding datasets directories:

• ICVL. Arad, B. and Ben-Shahar, O., 2016. Sparse recovery of hyperspectral signal from natural RGB images. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part VII 14 (pp. 19-34). Springer International Publishing.
• CAVE. Yasuma, F., Mitsunaga, T., Iso, D. and Nayar, S.K., 2010. Generalized assorted pixel camera: postcapture control of resolution, dynamic range, and spectrum. IEEE transactions on image processing, 19(9), pp.2241-2253.
• KAUST. Li, Y., Fu, Q. and Heidrich, W., 2021. Multispectral illumination estimation using deep unrolling network. In Proceedings of the IEEE/CVF international conference on computer vision (pp. 2672-2681).
• Harvard. Chakrabarti, A. and Zickler, T., 2011, June. Statistics of real-world hyperspectral images. In CVPR 2011 (pp. 193-200). IEEE.

To convert the hyperspectral images into a format that is compatible with the software, use the hsi2envi.py script.

Note

The ICVL-MATLAB directory is intended for the downsampled images with .mat extension, whereas the ICVL-ENVI-RAW is intended for the original output of the hyperspectral camera with paired .hdr/.raw files. The directory ICVL-ENVI should be populated by manually downsampled images using the resample_icvl.py script.

To prepare the datasets for training of spectral reconstruction models, use the synthesize_cave.py, synthesize_icvl_envi.py, and synthesize_icvl_matlab.py scripts (other datasets were only used for software testing).

License

MIT, Copyright © 2025, Daniils Buts