An automatic personalized photo recommender system based on learning user preferences

This paper presents an automatic personalized photo recommender system which recommends photos from a large collection. Our proposed system recommends photos based on user-preferences about aesthetics and basic quality features of the photo. A large dataset is put together, which is used to collect user-preferences. A random forest based learning system has been employed to learn the user preferences about different image quality features including aesthetic features. The system is validated using a part of the collected user preferences as ground truth and it has been compared to the baseline of random selection of photographs. Our automatic system significantly outperforms random selection, which shows the usefulness of our proposal especially when the collection of photos is manually unmanageable.