Automatic license plate detection by fast stroke filter

Recognition of license plate or detection of license plate is one of the popular applications in surveillance system, but to capture a plate in the natural scene becomes a difficult problem because of insufficient light and dirty plates. In this paper, we proposed a system that combined the fast stroke filter (FSF) to quickly extract and recognize the plate in traffic image. The extracted broken plate characters have been reconstructed by using the Modified Connect Component Analysis (MCCA) and grouped into one or more plates by using High-level Information Grouping (HLIG). In experiment, our system is being tested with some image databases, which are captured from different natural scenes; furthermore, the result shows that the proposed system reaches 89% detection rate with 2% false alarm rate and 88% detection rate with 4% false alarm rate to one or more plates in image.