The Research On Preprocessing Technology And Character Segmentation Of Vehicle License Plate Image

Automatic License Plate Recognition(ALPR) acts as an important role in Intelligent Transportation System(ITS). Usually, ALPR can be composed of three parts, including plate extraction, character segmentation and character recognition. However, due to the interference of non-uniform illumination, shadow and background noise, character segmentation still meets great challenge, which attracts us to do some research. The main methods of the proposed frame are as below:Using the high density of vertical edges and similarity of edges' gray value pair, the horizontal tilt correction algorithm focuses on excluding background interference. After vertical edges' filtering and horizontal connecting, effective columns of character area are selected. The center coordinates of every effective column are used to fit the line by least square method. Plate with tiny tilt angel and negative result given by tilt judging method will not be rotated.The color pattern judging step uses both color features and character texture features. By analyzing the probable color type of plate region, the plate with obvious color features will be judged by quantized color histogram. And the plate with low saturation will be judged by binarization and morphology process according to the consistency of stroke width. This method demands little of binarization and eliminates the effects of none-plate area.The proposed self-adaption binarization method gets result by selecting two algorithms, making use of gradual change of illumination and empirical rate of character area. First, the plate image is threshed by method based on Otsu and Bernsen to against non-uniform illumination. Then, some conditions are used to judge whether to binarize it by color clustering to against the influence of character shadow.Taking advantage of color consistency of plate region background and arrangement mode of character, our character segmentation method combines projection analysis and template matching algorithm. Firstly, get the valleys of the binary image's vertical projection. Color consistency is used to estimate the right-left border and find the big separator. Then, with the constraint of projection valleys and reference of the big separator, the best matching parameters of the designed template are decided by maximum variance classify rule. Finally, the separate location is adjusted by projection analysis to get accuracy result.The segmentation precision of 700 images is 97.28%, which demonstrates the robustness of the proposed methods against non-uniform illumination, background interference and other noise.