Abstract
License plate recognition (LPR) system is an import part of intelligent transportation system, which is mostly applied to electronic payment system, such as non-stop toll collection in highway, non-attended parking fee payment, and multi-use payment. LPR system has been developed for many years. Under outdoor working conditions, various illumination and non-stationary backgrounds may be the two factors that mostly affect the quality of scene images and complexity of the techniques. In this study, as few constraints as possible in the working conditions are considered. The proposed LPR algorithm consists of four modules: license plate locating, color classifying, characters segmenting, and characters recognizing. Research achievements and innovations are as follow:
Texture of license plate has fractal features. According to these features, an algorithm for license plate extraction is proposed, which is based on directional fractal parameters. Robust of algorithm is analyzed and two main parameters, template size and comparative pixels distance are discussed, and this is useful to find license plate candidates in complex scene.
HSV color model is introduced in license plate color classifying. Gray scale distribution feature in H map and S map is studied for all kinds of license plate. Further, combined with histogram features analyses, a license plate color-classifying method based on HSV model and histogram features is proposed.
Rotating projection method is introduced to find the orientation of an inclined plate. After normalizing the inclined plate by bilinear interpolation, location map is converted into a binary image, and then projected vertically. In projection map, with the help of license plate geometrical constrains, license numbers are segmented into isolated characters.
Character deformation, such as shear and skew, often affects recognition. A proposed method for recognizing distorted characters is constructed from a Zernike-moments-based approach, which extracts rotation-free property of character.