Visual Tagging Through Social Collaboration: A Concept Paper

Abstract

Abstract. Collaborative tagging has grown on the Internet as a new paradigm for web information discovering, filtering and retrieval. In the physical world, we use visual tags: labels readable by smartphones with cameras. While visual tags are usually related to a web site address, collaborative tagging, instead, provides updated, recommended information contributed and shared by users. In this paper we investigate the combination of collaborative tagging systems with visual tags. We present a prototype of a semiautomatic system generating visual tags which gather information from collaborative tagging. The user can interact with a list of relevant tags (built by clustering closely related tags) that can be further encoded in a visual tag, according to user's preferences. The user experience is enriched by retrieving multimedia content linked to the selected tags, present on the web. We finally show a case study illustrating our approach.

Visual tag authoring

“…The use of location-based metadata has also been studied in [9] where the presented system allows: a) to share tagged images (photographs and related tags) and b) to collect tag patterns present in geo-tagged photos. In our previous work [10] we introduced the prototype of a semiautomatic system generating visual tags by gathering information from collaborative tagging. The user authoring the visual tag interacted with a list of given tags (key-words obtained employing Flickr's clustering service) and selected tags which best capture his/her needs.…”

“…As seed tag we mean to search for images starting from a given tag or area. As we suggested in [10], these tags can be further encoded into a single visual tag. Barcodes are an example of visual codes: machine readable labels easy to generate and print.…”

“…Differently from [10], instead of employing the Flickr clustering service we developed our own clustering application. Starting from a given tag (i.e.…”

“…The representativeness of a photo with regard to a given tag has been computed through the "most interesting" function as provided by Flickr API. The main idea behind our approach is to generate a set of images (selected by the feature vector ) for automatically creating a visual tag describing a location of interest [10]. Through the system interface the authoring user can select different information about the cluster (such as tags in the feature vector along with their score, the number of distinct users and the number of photo shots) to be encoded in the visual tag (the QR Code).…”