An Automated Pottery Fragment Classifier for Archaeological Studies

'. t. •'...,/, ,.,.•::•! :• ,;.'•,;!•••..'',//( l/iis paper we propose an image-based pottery fragment identifier and classifier. We have successfiilly developed an automated, search engine system for pottery and potteiy fragment images. This system is designed to assist archaeologists and students in identifying pottery shape, color convention, and other relevant information quickly and accurately The purpose of this automated pottery fragment classifier is to make the matching task easier for archaeologists by providing them with a graphical user interface for matching and classification. We present several image retrieval and computer vision techniques and describe their applications within the domain of archaeological studies by utilizing a large digital library of pottery photographs. Potteiy shapes .tuch as amphora, kylix. hydria and lekythos. and pottery schools such as Red Figure. Black Figure, and Wliite Ground are identified with shape and color-based image retrieval techniques, respectively. The system analyses and compares extracted features to determine the five closest database images, and then presents them to the user for final decision. This is the first potteiy study to combine the two different techniques regional property measurements and color-based image retrieval lo identify multiple characteristics of an unknown pottery image or a pottery fragment. The database contains one hundred sixty pottery images obtained from online digital libraries. Experiments on identifying the correct pottery shape, school, and cropped images yielded approximately 98% accuracy.