A Comparison of Multispectral Transforms

Abstract

This research compares the results of five multispectral transforms applied to images of carbonized scrolls to determine which transform creates the best readable image. These transforms are vector quantization with principal components analysis, noise subspace projection, interference-and-noise-adjusted principal components analysis, convex cone analysis, and penalized discriminant analysis with principal components analysis. One approach to interference-and-noise-adjusted principal components analysis called signal-to-interference-plus-noise-ratio-based principal components analysis had the highest readability score according to a subjective judgment of 30 randomly selected individuals. However, convex cone analysis created a resultant image with the highest signal-to-noise ratio.