Parallel neural networks for multimodal video genre classification
Multimedia Tools and Applications
| Hi-index | 0.00 |
TV stream structuring is the first step in the description process of telecasts for TV archives holders like the French National Audiovisual Institute. It consists in breaking down automatically the stream into telecasts and advertisings. One can think TV stream structuring consists in a simple alignement of the TV guide (for example Electronic Program Guide or TV magazines) on the stream. But our study shows that only 34% of telecasts appear in program guides. In average, 2 hours per day, composed of short telecasts and advertisings, are missing. We propose in this article a way to model and predict TV schedules (one day or one week of telecasts on a channel) by using a new extension of Markov models and regression trees. The goal is that the improved schedules obtained by the prediction can be aligned on the stream.