Modeling and clustering of photo capture streams
Ullas Gargi · 2003
We present results of statistical modeling of consumer photographic media capture behaviour based on timestamp metadata. Working with real data sets, we show that it is bursty and is not well described by a Poisson model. We show that it is in fact a fractal process, with fractal dimension characteristic of the consumer. We also present an algorithm for temporal segmentation of consumer media capture streams into event clusters based on timestamp metadata, an algorithm for minimal labeling of such clusters, and a method for visualizing the results of such clustering on highly bursty data. Results of using software implementations of the algorithms to organize and to browse consumer photo collections within a multimedia information management system are presented.