Audio-visual video classification system design: For Arabic News domain

Abstract

There are many research initiatives tackling automated multimodal video classification, however very few of them are targeted towards classifying Arabic news-related videos. In light of the vast proliferation of raw digital Arabic data, specifically videos, over the internet, uncategorized and unused, we propose a new system to tackle this problem. The proposed system design consists of visual features extraction and classification, combined with audio-based event classification, as well semantic-content processing. Results are to be combined and documented using multimedia classification fusion techniques. We also propose to develop a new Arabic dataset based on news channel videos as well as raw videos from various online sources for testing and evaluation.