We present a near realtime text tracking system capable of detecting and tracking text on outdoor shop signs or indoor notices, at rates of up to 15 frames per second (for generous 640x480 images), depending on scene complexity. The method is based on extracting text regions using a novel tree-based connected component filtering approach, combined with the Eigen-Transform texture descriptor. The method can efficiently handle dark and light text on light and dark backgrounds. Particle filter tracking is then used to follow the text, including SIFT matching to maintain region identity in the face of multiple regions of interest, fast displacements, and erratic motions.
|Translated title of the contribution||A framework towards real-time detection and tracking of text|
|Title of host publication||Second International Workshop on Camera-Based Document Analysis and Recognition (CBDAR 2007)|
|Publication status||Published - 2007|
Bibliographical noteOther page information: 10-17
Conference Proceedings/Title of Journal: Second International Workshop on Camera-Based Document Analysis and Recognition (CBDAR 2007)
Other identifier: 2000721