Research On License Plate Recognition Based On AdaBoost Algorithm

Coal output and transport volume control system are of great significance to solve the challenging situations caused by coral industry particularity and to realize modernization of coral industry management On basis of coal transport volume control system design, this paper researches on license plate recognition which is part of coral transport volume control system, that is, automatically recognize license plate by digital and image processing technique. The technique and methods explored in this paper will lay a solid foundation for further researching and application.In this paper, the key technique to recognize license plate is explicated, detailing in license location, character segmentation, feature extraction and character recognition. At the license location section in this paper, considering the coral field as a particular background, this paper puts forward a combined way to locate license plate, which combining the length of connected components chain, edge density, and color features. The whole process goes as follows:Firstly, binary processing the image by Niblack algorithm. This method works well in conditions that the license plate is not evenly illuminated. Then analyzing connected components of binary processed image and matching connected components chain. Lastly, accurately locating the license plate through combining the length of connected components chain, edge density and color features.In the section of character segmentation, a way for improving AdaBoost algorithm is put forwarded to deal with overfit of AdaBoost. It is a way through adjusting weight based on distribution of weight value. First the weight of a certain training session is divided into segments evenly, then the number of weight value in each segment is figured out and its proportion accounting for the whole value number is calculated. The final weight value is got by multiplying the proportion by weight value. The simply classified data is iteratively trained in the improved algorithm and emulation experiment is conducted. The result proves that this way significant improve the overfit of AdaBoost.