| Video indexing and annotation are important steps in multi-media document understanding and information retrieval. Because of the high variability of video types and the accuracy requirement of indexing, currently index construction is often done manually. Manual indexing of video documents is expensive and time consuming. Therefore, automatic video indexing methods are necessary. This thesis presents a novel machine learning based approach for automatic indexing and annotation of lecture videos based on specially designed text features and deep neural networks. By indexing video content, we can support both topic indexing and semantic querying of multimedia documents. Video annotation is produced by the association of video segments to different visual objects and other meta data. The proposed system offers indexing and annotation features that are suitable for research, teaching or learning. As part of the annotation, we introduce a novel mathematical formula detection approach. It detects and locates mathematical formulas in document images. The proposed automated lecture video indexing and annotation approaches make lecture videos more convenient to search and navigate and help users locate and access lecture content more efficiently. |
