A comparison of methods for image classification of cultural heritage using transfer learning for feature extraction

Abstract

Image recognition and classification in the domain of cultural heritage is a complex task that usually requires good image quality and a large dataset. Small dataset is an apparent problem when data availability is limited, often resulting in poor classification performance. The aim of this study is to analyze and compare the performance of four machine learning algorithms, namely random forest, multilayer perceptron classifier, Naïve Bayes, and decision tree, that are used to classify features extracted using eleven pre-trained deep learning architectures from a small set of images representing cultural heritage. The findings imply that random forest and multilayer perceptron classifiers are the most appropriate for the task of image classification of small cultural heritage datasets, as they obtained the highest performance compared to the other two algorithms, particularly when DenseNet121, EfficientNetB0 and NASNetMobile architectures are used for feature extraction. In addition, for the classification of features extracted using NASNetMobile all the four utilized machine learning algorithms obtained high accuracy ranging from 88.89 to 95.56%.