Ternary classification for image aesthetic assessment using deep learning

摘要

As the number of digital photos increases, the need for image aesthetic assessment is increasing in various applications to provide improved user satisfaction. Most existing studies have considered binary classification to determine whether an image has a high- or low-level of aesthetic quality. However, the binary classification has a limitation in that when an image is classified incorrectly, users experience a large gap between their perception and the predicted result. To reduce the gap, we propose ternary classification-based image aesthetic assessment. Through experiments using popular classification deep learning models, we show the advantages of the ternary classification over the binary classification.