Research On License Plate Recognition In Complex Background

License plate recognition system is the most important part of modern intelligent transportation system.It has been widely used in all kinds of unrestricted outdoor scenes and has great application value and research value.The current license plate recognition system has a high recognition rate in certain fixed traffic jam places and expressway toll stations under good lighting conditions,but it still cannot fully solve the problems of license plate recognition under the complex background of fog weather,background interference,sewage interference,license plate tilt and so on.How to achieve the balance between high accuracy and high efficiency of license plate recognition in the complex background has been the focus of many scholars in the field of license plate recognition in recent years.In order to improve the adaptability of license plate recognition under complex background,this paper is based on different recognition algorithms,and the main research work is as follows:In terms of fogless pretreatment and recovery of license plate image,three fogless filtering algorithms were compared and analyzed for license plate image in foggy weather formed by air particles,and the fogless optimization algorithm based on trilateral filtering in YCrCb space was selected as the main algorithm to prevent the influence of atomization in the air on license plate positioning stage.Secondly,in the stage of license plate image localization,a license plate localization algorithm based on improved Canny operator and mathematical morphology was proposed.The filter in Canny edge operator is replaced by a hybrid filter,and the best threshold is obtained based on the edge processing based on linear enhancement and the maximum inter-class variance method.The morphological processing is more suitable for license plate positioning.Then,in the segmentation stage of license plate characters,the structural ideas of FCM algorithm and WFCM algorithm were analyzed.In view of the fact that the algorithm did not consider the relevance of spatial information,the covariance factor was introduced to improve the algorithm to achieve the accurate segmentation of license plate characters.Experimental test results showed that the segmentation effect of license plate was improved.Finally,in the stage of license plate recognition,a license plate recognition algorithm combining HOG feature and SVM algorithm is used to recognize license plate images.Through the test results of license plate data set,it can be concluded that the license plate recognition system designed in this paper can meet the requirements of high license plate recognition success rate under complex background.