SHIATSU: Semantic-Based Hierarchical Automatic Tagging of Videos by Segmentation using Cuts

The recent dramatic widespread of multimedia content over the Internet and other media channels (such as television or mobile phone platforms) points the interest of media broadcasters to the topics of video retrieval and content browsing. Semantic indexing based on content is a good starting point for an effective retrieval system, since it allows an intuitive categorization of videos. However, the annotation process is usually done manually, leading to ambiguity, lack of information, and translation problems. In this paper, we propose SHIATSU, a novel technique for automatic video tagging which is based on shot boundaries detection and hierarchical annotation processes. Our shot detection module uses a simple yet efficient algorithm based on HSV histograms and edge features. The tagging module assigns semantic concepts to both shot sequences and whole videos, by exploiting visual features extracted from key frames. We present preliminary results of our technique on the Mammie platform video set by proving its effectiveness in real scenarios.