Multimodal Genre Analysis Applied to Digital Television Archives

Automatic genre classification is a simple and effective solution to describe semantic properties of multimedia data. In this paper, a method to classify the genre of TV programmes is presented. In our approach, four multimodal vectors, including both low-level perceptual descriptors and higher-level, human-centred features are employed. These vectors serve as the input for a parallel neural network system that performs classification of seven video genres. The experiment results confirm the effectiveness of our method, reaching a classification accuracy rate of 96%. In addition, the results show the correlation between the analysed genres and the classes of the extracted descriptors, demonstrating their effectiveness in explaining what we call "the multimodal essence" of the genres.