Incremental statistical geo-temporal structuring of a personal camera phone image collection

This paper makes a proposal for automatically organizing the personal image collection that would be collected from a mobile phone equipped with a digital camera. Doing so, it attempts to address emerging needs from this rapidly developing device family. Having sketched user needs, we underline the interest of temporal and spatial meta-data in the chosen context. Collection organization is then formulated as an unsupervised classification problem, in both space and time. A criterion and an estimation procedure are proposed, based on the statistical integrated completed likelihood criterion, providing effective solutions to model complexity determination, non-Gaussianity of clusters, and incrementality. For further summarization, a technique for fusing the temporal and geolocation-based partitions is finally put forward, to ease browsing of the image collection along a single dimension.