Unsupervised Clustering of Hyperspectral Paper Data Using t-SNE.

Area of Science:

• Forensic Science
• Document Analysis
• Imaging Technology

Background:

• Document forgery analysis traditionally involves examining paper and ink.
• Hyperspectral Imaging (HSI) is a non-destructive technique offering rich spectral data for forensic analysis.
• HSI captures numerous narrowband images across the electromagnetic spectrum, surpassing conventional imaging.

Purpose of the Study:

• To evaluate the effectiveness of the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm for classifying hyperspectral paper data.
• To assess the t-SNE algorithm's utility in forensic document analysis, specifically for paper authentication.
• To compare the performance of t-SNE with traditional Principal Component Analysis (PCA) and k-means clustering.

Main Methods:

• Development of a hyperspectral dataset comprising various paper samples.
• Application of the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm to the hyperspectral paper data.
• Visual and quantitative evaluation of the clustering quality achieved by the t-SNE algorithm.

Main Results:

• The t-SNE algorithm demonstrated significant discrimination power in classifying hyperspectral paper data.
• Visual assessment confirmed the superior clustering capabilities of t-SNE.
• Quantitative evaluation further supported the exceptional performance of t-SNE over traditional PCA with k-means clustering.

Conclusions:

• The t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm is highly effective for the classification of hyperspectral paper data in forensic document analysis.
• HSI coupled with t-SNE offers a powerful, non-destructive method for establishing document authenticity.
• This approach provides enhanced classification results compared to conventional methods, aiding in forgery detection.