Yang, Haojin; Sack, Harald; Meinel, Christoph
International Journal of Multimedia Processing and Technologies (JMPT)
Texts displayed in lecture videos are closely related to the lecture content. Therefore, they provide a valuable source for indexing and retrieving lecture videos. Textual content can be detected, extracted and analyzed automatically by video OCR (Optical Character Recognition) techniques. In this paper, we present an approach for automated lecture video indexing based on video OCR technology: first, we developed a novel video segmenter for the structure analysis of slide videos. Having adopted a localization and verification scheme, we perform text detection secondly. We apply SWT (Stroke Width Transform) not only to remove false alarms from the text detection stage, but also to analyze the slide structure further. To recognize texts, a multi-hypotheses framework is applied, that consists ofmultiple text binarization, OCR, spell checking and result merging processes. Finally, we have implemented a novel algorithm for extracting lecture structure from the OCR-transcript by using geometrical information and text strokewidth of detected text lines. We use both segmented key frames and extracted lecture outlines for the further videoindexing. The accuracy of the proposed approach is proven by evaluation.