SVD-clustering, a general image-analyzing method explained and demonstrated on model and Raman micro-spectroscopic...

Area of Science:

• Multidisciplinary image analysis
• Data science
• Spectroscopy

Background:

• Traditional image analysis methods struggle with complex, high-dimensional data.
• Extracting subtle variations and spatial distributions from measurement data requires advanced techniques.

Purpose of the Study:

• To present a universal image analyzing method, SVD-clustering, for enhanced data interpretation.
• To demonstrate the application of SVD-clustering in revealing minute alterations in real-world samples.

Main Methods:

• Singular Value Decomposition (SVD) factorization to obtain amplitude vectors (V_i) and basis-spectra (U_i).
• Mapping V_i vectors to reconstruct spatial distributions of U_i basis-spectra.
• Extending SVD with clustering by using significant V_i vectors as coordinates in an n_e-dimensional space.

Main Results:

• SVD-clustering successfully reconstructs spatial distributions of basis-spectra, providing insights into higher-order deviations.
• Clustering applied to image points in an n_e-dimensional space effectively groups similar data.
• Demonstrated theoretical possibilities and limitations on models, and revealed subtle changes in mineral cation ratios and plant root cellulose structures in real samples.

Conclusions:

• SVD-clustering is a universal and powerful tool for analyzing diverse 2D and 3D image data.
• The method offers high spatial resolution for detecting minute alterations in samples.
• Applicable to any measurement dependent on an external parameter, broadening its scientific utility.