Abstract
Luster assessment stands at the crossroads of different fields and there is very few literature specifically dedicated to it. In a perspective of automating culture pearls' luster assessment, a way to extract features out of pearls' photographs is proposed and tested on a real dataset labeled by a human expert. After training, an SVM using these features can predict luster quality of new pearls with up to 87.3 % (± 5.7) accuracy. Moreover, it turns out that some of these features could be used for developing an objective luster quality control.