Video categorisation mimicking text mining

Abstract

With the rapid growth of online videos on the Web, there is an increasing research interest in automatic categorisation of videos. It is essential for multimedia tasks in order to facilitate indexing, search and retrieval of available video files on the Web. In this paper, we propose a different technique for the video categorisation problem using only visual information. Entity labels extracted from each frame using a deep learning network, mimic words giving rise to manage the video classification task as a text mining problem. Experimental evaluation on two widely used datasets confirms that the proposing approach fits perfectly to video classification problems. Our approach achieves 64.30% in terms of Mean Average Precision (mAP) in CCV dataset, above other approaches that make use of both visual and audio information.