Learning personalized expectation-oriented photo selection models for personal photo collections

Human expectations and practices are key aspects to consider when developing semi-automatic methods to select important photos from personal collections, e.g. for creating an enjoyable sub-collection for revisiting or preservation. The photo selection process (especially for personal data) can be highly subjective and the factors that drive the selection can vary from individual to individual. Thus, generic selection models might have limitations in meeting the different expectations and preferences of each user. In this paper, we propose a personalized photo selection model to assist users in photo selection, which adapts to their selection behaviors and preferences. Given an initial selection model trained on the available data, selection decisions done for new collections are acquired and the selection model is re-trained accordingly. Our experiments, based on real-world personal photo collections with overall more than 18,000 images, show promising adaptation capabilities of our personalized selection models.