Content-driven adaptation of on-line video

This work presents an on-line approach to the selection of a variable number of frames from a compressed video sequence, attending only to rules applied over domain-independent semantic features. The localization of these semantic features helps infer the heterogeneous distribution of semantically relevant information, which allows to reduce the amount of adapted data while preserving meaningful information. The extraction of the required features is performed on-line, as demanded by many leading applications. This is achieved via techniques operating on the compressed domain, which have been adapted to operate on-line, following a functional analysis model that works transparently over both DCT-based and wavelet-based scalable video. The main innovations presented here are the adaptation of feature extraction techniques to operate on-line, the functional model to achieve independence of the coding scheme, and the subjective evaluation of on-line frame selection validating our results.