An Improved Method of Video Text Detection

This paper presents an improved video text detection method. At first, Harris corners detection is fulfilled to the original region. After confirming the former three text frames, this algorithm gets a suitable candidate text region according to the characteristics of these text frames and only processes candidate text region in later corner detection in order to improve detection efficiency. Then, the obtained corners are calculated by row and a local threshold is gotten based on height and density of continuous corners rows to eliminate non-text corners rows. Finally, the center location feature is employed to make sure text region further and confirm the text frames. The experiment indicates that this method has relative high detection accuracy.