13 - 18 OCTOBER 2013, ATLANTA, GEORGIA, USA

Multimedia Pivot Tables

Contributors: 
Marcel Worring, Dennis C. Koelma
Description
Image collections are a tremendous source of information. Yet due to the semantic gap it is difficult to get access to their content, while at the same time it is difficult to properly employ their context such as tags and metadata. To move forward we propose a multimedia analytics solution. The most widespread and universally used analytic tools are spreadsheets, where a powerful feature is the possibility to generate pivot table reports. They provide flexible interactive summaries of the data along various dimensions. Pivot tables have been designed and are in use for structured data. Our goal is creating pivot tables for accessing collections of images, their content, tags, and metadata. This is a challenging task as automatic descriptors for image content are noisy, tags are numerous and subjective, and metadata can have many types. To tackle these challenges we present methods and visualizations for semi-interactively categorizing an image collection and from there design and develop pivot tables for such a collection.