Visual content summarisation for instructional videos using AdaBoost and SIFT

Research contributions in video retrieval field are rising to propose solutions for automatic understanding and retrieval of video content. The aim is to make the user able to retrieve specific video sequences in a large database, based on semantic information. In this paper, we process a special case of videos, instructional videos, where text presents very rich semantic information for understanding video content. Indeed, lecture videos are the source of information used in learning systems by educators and students for archiving and sharing knowledge. However, users usually have difficulties to access accurate parts in instructional videos. In our paper, we propose a method to summarise the visual content in instructional videos. For that, first, we segment the video into shots based on SIFT. Then, key frames which are rich in text and figures are extracted from each shot based on entropy measurement. These keyframes are classified using AdaBoost to eliminate non-text frames. The text content in the lecture video summary can be detected and recognised to identify keywords for indexing and classification.