License plate character segmentation based on stroke width transform

License plate character segmentation is an important step of license plate recognition (LPR). This paper proposes an approach to segment license plate character based on stroke width transform (SWT). The method detects some character candidates by SWT, reserves real character regions with cluster analysis based on horizontal position and the height of detected candidate regions, and fits two straight lines. According to the tilting degrees and locations of fitted lines, locate precisely position of a plate, and correct the rotation plate. At last, segment characters using connected component analysis and template matching. Experiments show that the proposed method can accurately segment license plate characters, has good adaptability to rotation, slightly fouling. It also saves to be difficult to adjust parameters with a general method to deal with different scale plates, meanwhile not necessary to preciously locate the position of plates before segmentation.