Multi-layered videotext extraction method

The importance of video content analysis and retrieval increases as storage capacity grows in consumer devices. Videotext detection is very important for video indexing and retrieval enabling features such as commercial detection, intelligent keyframe extraction, program boundary detection and others. In this paper we propose multi-layered videotext detection tailored to an MP EG encoding scheme. We approached the videotext detection from a perspective of scalability and flexibility for different platforms with varying resources. We propose a three-layered algorithm. The first layer works in compressed domain features such as macroblock type. The second layer works in semicompressed domain such as DCT coefficients. The third layer works in uncompressed domain, i.e. spatial domain. As the next layer gets implemented the complexity increases but so does the precision of the algorithm. On constrained platforms, only the first layer would be implemented. On high-end platforms all three layers could be implemented to enable a full suite of indexing and retrieval applications.