iVision HHID: Handwritten hyperspectral images dataset for benchmarking hyperspectral imaging-based document forensic...

Area of Science:

• Forensic Science
• Computer Vision
• Document Image Analysis

Background:

• Hyperspectral imaging offers unique data for document analysis beyond visible light.
• Existing datasets often lack real-world complexity for forensic applications.
• Handwriting analysis is crucial for forensic investigations and personal identification.

Purpose of the Study:

• To present a novel hyperspectral handwriting dataset for advancing document image analysis.
• To establish a benchmark for evaluating forensic analysis methods on hyperspectral document images.
• To facilitate research in writer identification, ink analysis, and demographic prediction from handwriting.

Main Methods:

• Collected hyperspectral images of handwriting from 54 individuals using 12 different pen types.
• Acquired data with a spatial resolution of 512x650 pixels and 149 spectral channels (478-901 nm).
• Included real mixed samples with varying ink ratios and writer combinations for forensic relevance.

Main Results:

• The dataset comprises 54 individuals' handwriting with detailed annotations.
• It contains 28 sentences per subject, including alphabets, digits, and diverse pen inks.
• The data supports tasks like writer identification, ink mismatch detection, and demographic prediction.

Conclusions:

• The presented hyperspectral handwriting dataset is a valuable resource for forensic document examination.
• It enables the development and benchmarking of advanced AI-driven forensic analysis tools.
• This dataset will drive innovation in automated handwriting recognition and forensic science.