Retrieving Amateur Video from a Small Collection

Research on digital video libraries has been done in extensive and expensive projects (e.g. Open Video Project [1], Físchlár [2], Informedia [3]). Small video collections have small budgets and cannot afford sophisticated techniques to put their material on-line. Though very basic digital video library features can be good enough for enlarging the access to rarely seen material, e.g. folklore films from the 1920’s to the 1990’s owned by the National Centre for English Cultural Tradition (NATCECT). This material is unique but rarely used as the archive opens few hours a week: digital access would make it widely available to scholars, students, and enthusiasts. Part of the NATCECT video collection was digitized (56 videos, 37 hours); movies were often of poor quality (recorded by amateurs and copied many times) and advanced video retrieval techniques could not be used; e.g. audio indexing via speech recognition was useless for silent or musical videos. Boundary detection was used to automatically generate summaries of the visual content in the form of sequences of images; the poor quality of the video source produced a high number of corrupted images requiring an initial selection. A short description of the video content and a set of archival information were manually added as metadata to support textual search.