Building Friend Wall for Local Photo Repository by Using Social Attribute Annotation

Abstract

In this paper, we focus on providing a novel image browsing and visualization experience for local photo repository. The so-called Friend Wall system solves two problems: (1) How to effectively classify the local images with respect to related social characters and events. (2) How to efficiently generate layout to compactly arrange many photos onto a single canvas. For the first problem, we propose a novel image annotation scheme by employing both of the image visual features and Metadata. Motivated by the observation that SNS (Social Networking Service) images, especially those come from the user’s acquaintances always contain rich information, we apply these images as our training set and explore their social attributes. In our definition, social attributes contain a set of intrinsic labels such as Who, When, Where, and What. To effectively arrange the photos on a single canvas, we proposed a binary tree based representation and fast algorithm for layout generation. Experiments show the examples of Friend Wall. The effectiveness of social attribute annotation is proved as well.