Research On Character Recognition Algorithm Based On Regular Extreme Learning Machine

Each RMB has a currency serial number.It can be used as a symbol of the identity.The financial sector can establish appropriate management mechanism to track the circulation of RMB.To research and implement RMB crown word number automatic identification system has far-reaching significance and prospect of wide for the financial mechanism in China.In order to identify the number of the crown and word,it can use the technology of digital image pattern recognition and feature extraction.Meanwhile it need combine with the digital characteristics and statistical characteristics of the crown number itself.The RMB image pretreatment includes crown word number character segmentation,feature extraction and character recognition.The main contents are as follows,(1)In the processing of the number of the results of the crown,it need compare with various two value of the algorithm,such as,OSTU algorithm,Bersen algorithm,based on the ratio of the two values of the algorithm is analyzed.For the background of the crown number of single and the characteristics of real-time and high accuracy of recognition,OSTU algorithm was selected as the crown word number of the two value algorithm.In the crown word number segmentation,by vertical division and horizontal segmentation algorithm and the crown word number of regional level and vertical area was determined.Based on a priori knowledge of the crown word number that background pixel point and the target pixel different histogram was between 80,the crown word number region was determined more accurately.(2)For the tilt correction of the characters,several tilt correction algorithm were compared.Hough transform to detect a straight line,to find the maximum tilt angle,and then choose the maximum tilt angle of the image correction.Compared with image tilt method based on least square method,the least square method is a kind of algorithm based on the least linear fitting error.This system chooses to use the linear tilt correction algorithm based on Hough transform.Feature extraction,corner feature,SURF feature,cross feature and thirteen grid features were compared.Thirteen mesh feature extraction is used in this paper.Thirteen mesh feature extraction has the small impact of the classifier,good independence and improve the recognition rate of characters(3)In this paper,we propose a new method for the classification of crown numbers based on genetic algorithm and extreme learning machine algorithm.The classifier based on support vector machine and BP neural network were compared.Comparison of experimental data shows that the algorithm is effective.