Video-text extraction and recognition
Teo Boon Chen, D. Ghosh, S. Ranganath · 2004
The detection and recognition of text from video is an important issue in automated content-based indexing of visual information in video archives. In this paper, we present a comprehensive system for extracting and recognizing artificial text from unconstrained, general-purpose videos. Exploiting the temporal feature of videos, an edge-detection-based text segmentation method is applied only on selective frames for extracting text from a video scene. Subsequently, a combination of techniques including multiple frame integration, gray-scale filtering, entropy-based thresholding and line adjacency graphs is used to enhance the detected text areas. Finally, character recognition is accomplished by using the character side profiles. Results obtained from experiments on uncompressed MPEG-1 video clips demonstrate the effectiveness of our proposed system.