抽象的な

Ranking of Landmark Images

B.Chandirika, S.Selvarani

A landmark is a recognizable natural or man-made feature used for navigation, a feature that stands out from its near environment and is often visible from long distances. In modern use, the term can also be applied to smaller structures or features that have become local or national symbols. This article presents an approach for mining landmark images from heavily contaminated image collections gathered from the Internet that achieves greater accuracy in traditional problems like landmark recognition. People, particularly tourists, are interested in viewing photos of landmarks across world-wide locations. To automatically identify landmarks from existing photo collections, methods based on only metadata, only content or a fusion of both are used. Certain computer vision techniques are also used to select iconic images for selected locations. Social tags with associated spatio-temporal information can be readily plugged into general techniques for landmark recognition. In this paper image mining techniques were applied on the accompanying metadata to determine independent ranking of images. In the first stage of processing, images are clustered based on global appearance descriptors, and the clusters are refined using certain geometric constraints. Using structure from motion techniques, the system then registers the iconic images to efficiently produce the different aspects of the landmark. To improve coverage of the scene, these aspects are subsequently extended using additional, non-iconic views. Photographs relevant to each landmark tag were retrieved and distinctive visual features were extracted from them. The results for landmarks include names, geographic hierarchy and its visual features