Leveraging Geo-Referenced Digital Photographs Thesis Introduction

As the photography world shifted from film cameras into digital cameras, computers now play a significant role in managing people’s photographs or, if you will, memories. Photos are stored, shared, searched and viewed — all in digital format. Managing large personal collections of digital photographs is an increasingly difficult task. As the rate of digital acquisition rises, storage becomes cheaper, and “snapping” new pictures gets easier, we are inching closer to Vannevar Bush’s 1945 Memex vision [2] of storing a lifetime’s worth of documents and photographs. At the same time, the usefulness of the collected photos is in doubt, given that the methods of access and retrieval are still limited. With digital photos, the opportunity to go “beyond the shoebox” is attractive, yet still not entirely fulfilled. One of the major hurdles for computer-based photo applications is the semantic gap. The semantic gap is defined by Smeulders et al. [8] as “the lack of coincidence between the information that one can extract from the visual data and the interpretation that the same data have for a user in a given situation.” Given perfect semantic knowledge about the photos, the task of organizing and retrieving from a photo collection would be made much easier. For example, if a system could automatically derive that a photo shows “Kimya drinking with Dylan at Robyn’s birthday, in New York”, this semantic knowledge could go a long way in helping users manage and retrieve from their collection. Sadly, current technology sometimes cannot even reliably detect that there are two people in the photo just described. The existing approaches towards photo collection management can be categorized into four main thrusts. First, there are tools to enable and ease manual annotation (e.g., [7]). These tools let the user rapidly enter semantic informa-