Using Expert-Derived Aesthetic Attributes to Help Users in Browsing Image Databases

Abstract

Image repositories often contain a large amount of metadata about their content. However many resources, such as photographs, have inherent aesthetic qualities that can be difficult to describe in a semantically consistent and usable manner, yet would be highly valuable for users in exploring large image repositories, such as Flickr. Automatically augmenting existing metadata with expert perspectives has the potential to give users a consistent aesthetic vocabulary to search and explore such repositories. SARA (Semantic Attribute Reconciliation Architecture) is a system that supports users to leverage domain expertise while searching for items in a metadata-rich domain. X2Photo is a tool built on SARA?s functionality to enable image searching based on a picture?s aesthetic characteristics and user-generated tags. This paper describes X2Photo in detail, the approach to augmenting visual media with expertise, and the evaluation results which reveal how semantically described aesthetics can support complementary search axes for image retrieval.