• Sellers, William
• Pormann, Peter
• Bhayro, Siam
• Arsene, Corneliu

Abstract

Multispectral/hyperspectralimageanalysishasexperiencedmuchdevelopmentinthelastdecade (Kwon et al. 2013; Wang and Chunhui 2015; Shanmugam and Srinivasa Perumal 2014; Chang 2013; Zhang and Du 2012). The application of these methods to palimpsests (Bhayro 2013; Pormann 2015; Hollaus et al. 2012) has produced significant results, enabling researchers to recover texts that would be otherwise lost under the visible overtext, by improving the contrast between the undertext and theovertext.Inthispaperweexploreanextendednumberofmultispectral/hyperspectralimage <br/>analysismethods,consistingofsupervisedandunsuperviseddimensionalityreductiontechniques (vanderMaatenandHinton2008),onapartoftheSyriacGalenPalimpsestdataset (http://www.digitalgalen.net).Ofthisextendedsetofmethods,eightmethodsgavegoodresults: threeweresupervisedmethods–GeneralizedDiscriminantAnalysis(GDA),LinearDiscriminant Analysis(LDA),andNeighborhoodComponentAnalysis(NCA);andtheotherfivemethodswere unsupervisedmethods–GaussianProcessLatentVariableModel(GPLVM),Isomap,Landmark Isomap, Principal Component Analysis (PCA), and Probabilistic Principal Component Analysis (PPCA). The relative success of these methods was determined visually, using color pictures, on the basis of whethertheundertextwasdistinguishablefromtheovertext,resultinginthefollowingrankingof the methods: LDA, NCA, GDA, Isomap, Landmark Isomap, PPCA, PCA, and GPLVM. These results were compared with those obtained using the Canonical Variates Analysis (CVA) method [6,7] on the same dataset, which showed remarkably accuracy (LDA is a particular case of CVA where the objects are classifiedtotwoclasses).Acomparisonwasalsomadewithadoublethresholdingandprocessing technique, developed as part of this project, which consists of the following: the darker overtext is carefullyidentifiedbythehumanoperatorandcoloredinwhite(threshold1),andthentheremainingundertext,whichisblackbutnotasblackastheovertextwas,ismadeevendarker (threshold 2). This last technique showed some initial encouraging results, but its success depends on the human operator selecting suitable cutting values. Figure 1 shows the results and a comparison of the different computational techniques applied to page 102v‐107r_B of the Syriac Galen Palimpsest data(http://www.digitalgalen.net)andforthepageobtainedwiththeultraviolet(365nm) illumination with green color filter (i.e. called CFUG). Ultimatelythechoiceoftechniqueisbasedonthepreferencesofthepersontryingtoreadthe manuscriptandtheprecisemakeupoftheoriginaldocumentbuteasyaccesstoanappropriate toolset is clearly highly desirable. Further work will consist of applying other reducing dimensionality techniques that enable the recovery of the undertext in palimpsests, as well as applying the above techniques to the rest of the Syriac Galen Palimpsest.

Topics

• impedance spectroscopy