On-the-fly fashion photograph recommendation system with robust face shape features

In this paper, we demonstrate a novel method for retrieving fashion photographs with robust face shape features, a challenging problem due to the different face orientations, variant face size, and variation in lighting environment. In addition, we provide a novel dataset for labeling eight face shape categories, to enable future research on face shape classification. We also present experimental results through the proposed face shape features to reach 94% accuracy of face shape classification. Finally, we demonstrate an on-the-fly prototype application for the face-shape-dependent fashion photo recommendation system.