Two-Dimensional Character Recognition Algorithms For Robots And Their Applications

Character is a tool for human communication,and it exists widely in all walks of life.With the advent of the Internet and information age,all walks of life are in urgent need of automatic character recognition.This paper chooses two typical scenarios: the recognition of letters and numbers on electronic chips in electronic manufacturing industry and the recognition of Chinese chess characters in Chinese chess robots,to explore the recognition algorithm for the whole scene of character recognition.A template matching recognition method based on pixel difference is proposed for the recognition of English letters and digits on the chip;a machine learning recognition method based on target contour and skeleton features is proposed for Chinese characters on chess pieces;and a template matching and skeleton recognition method are proposed for the low recognition rate of the whole chip caused by similar characters in English letters and digits.Fusion algorithm of machine learning.The work of this paper is as follows:(1)A template matching recognition method based on pixel difference is proposed for the recognition of letters and numbers in electronic chips.Firstly,histogram equalization is used to enhance image contrast,and Hough circle detection is used to locate the midline of the chip.Then,the ROI recognition region of chip characters is extracted by prior knowledge.Aiming at the problem of defective character segmentation,a character segmentation method based on character geometric features is proposed.In order to solve the problem of different sizes and inaccurate location of characters after segmentation,the minimum circumferential circle of characters is used to normalize and relocate.Finally,the character is recognized by template matching based on pixel difference.The experimental results show that the average recognition time of single character is 4.63 ms,and the recognition accuracy is 99.42%.This method realizes the fast and accurate recognition of single character in electronic chip.(2)Aiming at the problem of Chinese character recognition of chess pieces,a machine learning recognition method based on target contour and skeleton features is proposed.Firstly,Hough circle detection is used for rough location,color recognition,ROI region segmentation and mean binarization.Aiming at the problem of low precision of rough positioning,the minimum circumscribed circle of characters is used for secondary positioning.Subsequently,the outer contour and inner skeleton of character images are extracted,and Hu moments are extracted as feature vectors for the two images.Finally,it is fed into the SVM classifier of RBF kernel function for classification and recognition.Compared with other CNN methods,the average recognition time of single Chess is 20.42 ms,and the recognition accuracy is 99.92%.This method achieves fast and accurate recognition of chess.(3)Finally,aiming at the problem of low recognition rate(94.5%)caused by similar characters in English letters and numbers,a fusion algorithm of template matching and machine learning is proposed.The template matching based on pixel difference is used for one recognition,and the SVM classifier is used for the second recognition of similar characters.Experiments show that the average recognition time of the whole chip is 166 ms,and the recognition rate is 98%.Thus,the fast recognition of the chip can be realized and the high accuracy can be guaranteed.In this paper,two scenarios,the recognition of letters and digital characters on chips in electronic manufacturing industry and the recognition of Chinese characters on chess pieces in Chinese chess robots,are studied.A template matching recognition method based on pixel difference is proposed for the recognition of English letters and digits on the chip;a machine learning recognition method based on target contour and skeleton features is proposed for Chinese characters on chess pieces;and a template matching and skeleton recognition method are proposed for the low recognition rate of the whole chip caused by similar characters in English letters and digits.Fusion algorithm of machine learning.