Vehicle Plate Character Recognition Method Based On Improved Extreme Learning Machine

As the cars increasing rapidly around the world, vehicle plate recognitionhas became the key technology of the intelligent transportation system. Vehicle platerecognition technology plays an important role in raising traffic managementefficiency and saving manpower cost. The existing vehicle plate recognitiontechnologies are possessed of several deficiencies such as slow recognition speedand low accuracy. This paper studied a lot of image processing method includingvehicle plate location and character segmentation, and the vehicle plate recognitionmethod based on improved Extreme Learning Machine (ELM) algorithm has beenproposed, so that the rapid recognition speed and high recognition efficiency havebeen realized.In terms of vehicle plate location, the location method of projection fittingbased on the transformation characteristics of the plate grayscale is discussed deeply.Change the vehicle plate image to binary image based on the image grayscalecharacteristics and morphology processing, then extract candidate vehicle plateregion based on the projection fitting and regional analysis, remove thepseudo-region according to the prior knowledge and finally the target vehicle plateregion is obtained.In terms of vehicle plate segmentation, vertical projection method is mainlydiscussed in the paper. On the basis of the vehicle plate located accurately, removethe top and bottom borders based on the color jump method at the first. Then verticalprojection is applied to the vehicle plate image, and adaptive segmentation algorithmis proposed according to characters and projection features. Finally the left and rightborders can be removed effectively and the vehicle plate has been divided to singleimages.In terms of vehicle character recognition method, peculiarity of ELM algorithmis deeply analyzed in this paper. The selected process of ELM parameters has beenoptimized, so the improved ELM algorithm is proposed and applied to characterrecognition. According to the character features, two feature extraction methods for characters are discussed. Also according to the arrangement characteristics of verticalplate characters, three improved ELM classifiers including Chinese character、English character、English&digital are designed, each character is identified by itscorresponding serial number in vehicle plate. In addition, one function called refusalis designed in the English&digital classifier. Confusing characters can be chosen toartificial recognition, so that the recognition efficiency has improved and errorrecognition rate has decreased. Compared with the vehicle plate recognition methodbased on BP neural network, experimental results show that, the proposed methodcan guarantee the high recognition rate, also do the character recognition morequickly.the proposed recognition algorithm has gotten rapid recognition speed and highrecognition efficiency, therefore satisfying the practical application.At the end of the paper, modular system aiming at vehicle plate recognition isdesigned in the MATLAB. GUI based on vehicle plate recognition system isdeveloped and a complete system which is possessed of effective characteridentification function has been realized. The GUI of vehicle plate recognitionsystem based on proposed improved ELM can show the process of image processingsmoothly, and display the final character recognition result intuitively.