Fiber dye classification by spectral imaging

Abstract

Identification of colorants of artworks is of paramount importance in the context of museums and art galleries. We present a technique to discriminate the fiber dyes into natural or synthetic class using principal component analysis (PCA). Spectral imaging is used to measure the reflectance spectra of a variety of dyed wools in visible to near infrared (Vis/NIR): 400–1000 nm and short wave infrared (SWIR): 1000–2500 nm wavelength range. The full spectral range is segmented into nine partitions, and eigen vectors are extracted for each segment of training data. The same eigen vectors are used to compute the principal components (PCs) of training and test data. To classify test data, we successively increase the number of PCs and apply k-NN classifier to associate class label to the most similar training data. Results show over 93% overall accuracy with high precision in the range (1500–2500) nm using six PCs. By this technique natural Madder dyes can be classified from synthetic dyes with more than 98% accuracy.