Experimental evaluation and validation of different variants of principal component analysis on quaternion multispectral imaging

Avilés J.M.Q.; Singh P.; Pushparajesh V.; Talati B.J.; Juneja B.; Garg A. · Quaternion-Based Sparse Image Processing: Advances in Multispectral Processing (2025) · DOI: 10.1016/b978-0-443-29244-6.00010-7

Resumen

Quaternion multispectral imaging improves image processing by dealing with spectral statistics complexity through quaternion illustration, providing more accurate and insightful statistics through an aggregate of multispectral imaging generation and quaternion arithmetic. This chapter focuses on the experimental evaluation and validation of various principal component analysis (PCA) variants such as relaxed two-dimensional-quaternion PCAs (R-2D-QPCA) and two-dimensional-quaternion PCA (2D-QPCA) on quaternion multispectral imaging. This chapter employed the Indian Pine dataset, with data preprocessing involving Gaussian filtering and min-max normalization. Feature extraction was performed using scale-invariant feature transform and local binary patterns. In quaternion multispectral imaging, the R-2D-QPCA demonstrates superior performance over the standard 2D-QPCA across multiple evaluation metrics like improved accuracy, better structural similarity index measure, higher peak signal-to-noise ratio, and root mean squared error (3.5), execution time (9s). The study evaluated PCA variants in quaternion multispectral imaging, revealing distinct performance characteristics that impact image quality and interpretability differently.

Instituciones: Universidad Tecnológica del Perú