| License Plate Recognition technology is one of the important research topics in computer vision and pattern recognition technology in the field of intelligent transportation applications, it plays an key role in traffic management, used in Electronic police, bayonet and so on.License Plate Location, Character segmentation and Character Recognition is viewd the core of the License Plate Recognition technology.The main work as follows:(1) A significant license plate location method based on Lab color model is proposed.Firstly, complish an image preprocessing operation, the RGB image is converted to Lab image by using a lookup table, blue and yellow license plate areas are received with b component, black and white license plate areas with L component. Secondly, the color and texture significantly of a license Plate is used, combine with the method of relative threshold, overcome license plate location in uneven illumination and foggy environment. Combined with the characteristics of the shape of the license plate, the license plate area is located according to the method of morphological operations, connected domain analysis and so on.(2) The character segmentation and HOG feature extraction are realized in this paper. The character segmentation is implemented according to the same height and spacing features of the rear five consecutive characters.Because of the numerous edge features in the license plate character regions, the HOG method is chosen to operate the small local cell in images, which can guarantee illumination invariant of the image.The information of a image edge can well described by HOG method. It is apt to extract the feature of a license plate.(3) The second filter of samples. The K means clustering method of classic dynamic clustering method is used by the first filter of samples. The result of filter is the candidate samples. Considering that after the first screening, There may be a situation of all samples from individual clusters in one class is deviate from the standard sample. The second filter of samples is necessary. And get the final classic samples.(4) A fast KNN algorithm of character recognition is given.According to the characteristic of spatial distribution of the feature vectors is concentrated in the same class and it is discrete segment in different classes.So, it is just compute k nearest neighbors which from the spatial domain of character feature to reduce the calculation significantly. (5) A method in Optical Character Recognition of secondary recognition of the similar characters is given. Two similar characters is recognized once more. Firstly, To find out the similar characters confused. Secondly, the sensitive area of similar characters were analyzed and extracted. The characters is recognized twice. |
