2017 2 nd International Conference on Artificial Intelligence and Engineering Applications (AIEA 2017) ISBN: 978-1-60595-485-1 An Engraving Character Recognition Based on Machine Vision WANG YU, ZHIHENG WU, HONGBIN LIU, QIYU CHEN, XIANYUN DUAN, JUEXIAN MO, JIGANG TONG, FEI LIAO and QINGLIN LIN ABSTRACT In order to solve the problem of low efficiency, low quality and false detection of metal surface engraving character detection, we proposed a kind of engraving character recognition system based on machine vision. This system can identify and detect through the image acquisition of engraving area, and a series of mathematical and morphological treatments of binary image. It can extract the contour using optimized Laplace edge detection operator and combining with the convexity characteristic of engraved words. The system successfully carried out the experiment of engraving character recognition and it can detect the character recognition with high efficiency and high qualified rate. It has realized the automation of engraving character recognition test production line, and it has certain application value and promotion significance. KEYWORDS Machine vision; Engraved words; Identification detection; Edge detection. INTRODUCTION In recent years, with the rapid development of equipment manufacturing in China, the quality of metal surface in the manufacturing and mechanical sectors should be higher and higher. However, the identification and detection of metal surface engraving characters are more stringent. With the development of machine vision technology, all kinds of character recognition tests are widely used in industry. Word recognition technology is widely used in virtual reality, image search, human-computer interaction, unmanned driving, industrial automation, robot, license plate recognition and other fields [1]. Character recognition systems according to the object can be divided into the following three categories: handwriting recognition, print text recognition and scene character recognition in which the identification of metal parts in the industrial machinery field are included [2]. At present, the word recognition technology has made great progress, and the accuracy of the word recognition has been improved continuously [3], the recognition rate of off-line handwritten Chinese Wang Yu, y.wang@giim.ac.cn, Zhiheng Wu, Qiyu Chen, Xianyun Duan, Juexian Mo, Jigang Tong, Fei Liao, Guangdong Intelligent Manufacturing Research Institute, 650000 Guangdong GuangZhou, China Hongbin Liu. School of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650504, China Qinglin Lin. Guangzhou National Intelligent Power Technology Co. LTD Guang Gong Guang Zhou, China 95
Character recognition system and rule writing offline handwritten Chinese character recognition system recognition rate has reached more than 99% [4]. Character recognition accuracy can reach a higher level, however, if you use software on a PC, it can be slow to identify, and it's not portable. In practice, We often need portable text recognition devices with small size and high recognition speed. From the current situation, the foreign research results are more than that in China in terms of word recognition system implementation. Our country also has some research on the implementation of the word recognition system, and obtained good results [5-7]. Some research institutes in China identify the text of offline handwriting by sorting and extracting eigenvalues of characters [8]. Some college teams use the grid features of words to extract identification [9]. But in some certain scenarios the word recognition rate is far from reaching the required requirements, especially in the mechanical manufacturing area the recognition success rate of some large metal parts surface special engraving word recognition detection cannot meet the detection requirements. Therefore, We designed an engraving character recognition system based on machine vision. Though a series of pretreatment performed on the image of the engraving text area, using convexity characteristic of engraved words, and extracting the contour using optimized Laplace edge detection operator, we realized the engraving character recognition. On the basis of rapid detection and recognition, it is efficient, and it costs less than other testing devices. OVERALL FRAME DESIGN OF THE SYSTEM The whole system consists of two parts: the image acquisition part and the image processing part, and the two parts are connected through the gigabit network connection. The overall frame diagram is shown in figure 1. Firstly, the CCD industrial camera in the image acquisition module collects the image of the engraved area of the metal surface.then the image is transmitted to the image processing module by the network port communication between the computer and the identification software system. Finally, we can extract the contour and identify the detection. The light source controller can adjust the light intensity according to the size of the surface of the metal parts, and ensure that the CCD camera can obtain a high quality image. Figure 1. Overall frame diagram of the system. 96
OVERALL SOFTWARE AND HARDWARE DESIGN OF THE SYSTEM hardware mainly includes light source, CCD industrial camera, image acquisition card and industrial control machine. To choose the appropriate light source can be a good inhibition of environmental noise, improve signal-to-noise ratio, character sculpture convexity feature contour is more obvious in the image, maximize the contrast of the characteristics of carving word. The light source of the system is selected as a parallel source of white light, AFT - RL12068W LED light source. See figure 2. CCD industrial camera is better than other industrial cameras in the aspect of image transmission performance, such as small volume, high reliability. It is critical to choose the appropriate CCD industrial camera for the whole vision system. Considering the characteristics of the size and area of the surface engraving of metal parts and the more complete specifications of the engraving fonts, the system camera selects a 5-megapixel model as the MV - EM510M CCD industrial camera. See figure 3. In combination with the practical operating conditions of the software system, We selected Yan Hua ARK embedded industrial control machine, and use image acquisition card with C interface. So that we can not only ensure the stability of running of the system software on the production line, but also improve the reliability the reliability of the system on the surface engraving of metal parts. SYSTEM SOFTWARE DESIGN software framework design The system software processing flow is shown in figure 4, mainly including CCD camera, image acquisition, image processing and result display. After the optical fiber CCD sensor has been triggered, the camera begins to collect the image of the engraved area of the metal surface and convert image information to digital image information. Then there is a series of mathematical morphology denoising and filtering processing. Finally we can engrave word contour extraction, detection and recognition through edge detector. The qualified results of the test are qualified, and the unqualified results are not qualified. Figure 2. Light source material. 97
Figure 3. CCD industrial camera. software framework design Figure 4. software block diagram. The software interface is run on PC, and the software development environment of VS2010 is adopted. By optical CCD sensor trigger, the CCD camera obtain the engraving on the surface of the metal image, and identify and detect in real time, feedback the results of detection to the control end, and display the results in the software interface. 98
Figure 5. software interface. TEST AND ANALYSIS In the engraving character recognition test automation production line, It is very important to identify the surface engraving of metal parts quickly and accurately and to recognize and detect continuously and efficiently. After the optical fiber CCD sensor has been triggered, the camera begins to collect the image of the engraved area of the metal surface, recognition software system analyzes the image of the transformation and finally makes the contour extraction and recognition detection, which is the automatic identification and detection of the metal surface engraving words. A large metal stamping piece with the same length of the same length is detected in the identification detection system. The comparison test with the manual test is shown in table 1 in the case of the same number of metal parts. Detection method The length of Engraved font Table 1. Test data. Qualified Error rate% Unqualified Error rate% 994/1000 0.6 98/100 2 Artificial 981/1000 1.9 89/100 11 995/1000 0.5 99/100 1 Artificial 982/1000 1.8 84/100 16 995/1000 0.5 95/100 5 Artificial 986/1000 1.4 85/100 15 997/1000 0.3 98/100 2 Artificial 988/1000 1.2 86/100 14 993/1000 0.7 98/100 2 Artificial 982/1000 1.8 83/100 17 995/1000 0.5 94/100 6 Artificial 980/1000 2.0 85/100 15 99
From the results of the test, it can be seen that the system and the artificial simultaneous detection of the same engraving font length, the same number of products with the same number of products, the system detection error rate is within 1%, and is stable, however, the artificial error detection rate at about 2%, and unstable, system testing same engraving fonts length, the number of unqualified products surface engraving word metal product, system error detection rate within 6%, artificial at about 15%. Prolonged artificial detection can produce fatigue and affect detection rate. However, the system test is stable and the error detection rate is low, which ensures the qualified rate of the surface engraving metal parts to detect the production line. SUMMARY The design is based on the machine vision of the engraving character recognition system, which has the identification and detection of the metal parts of different engraving fonts. The detection system is faster than manual detection in the fusion of the fiber CCD sensor and CCD camera. The detection error rate is within 1%, within the margin of error, and it is stable. At the same time, the detection system can detect the engraving characters of different lengths of metal surface. The test results show that the detection system can effectively carved words on the surface of the metal for high efficiency, high qualified rate of the recognition test, which realized the automation of engraving character recognition test production line. The detection system has a wide application value and broad market prospect. ACKNOWLEDGMENTS 1. Partially supported by Guangdong province science and technology project (NO2016A010106005, 2015B090901007, 2015B010918001, 2016B090918100) 2. Partially supported by Dongguan production and research project (NO2015509109209) REFERENCES 1. Huang Pan. Natural scene text recognition based on deep learning [D]. Hangzhou, University of Zhejiang, 2016. 2. Li Lei. Study on text recognition based on artificial intelligence machine learning [D]. Chen Du, University of Electronic Science and Technology, 2013. 3. Longjun Yu, Xiangqun Zhang, Jiangqing Xue. A text recognition system based on FPGA [J]. Modern Electronic Technology, 2006, (11): 103-105. 4. Xiaoqing Ding. Review of Chinese character recognition research [J]. Electronic Journals, 2002, 30(9): 1364-1368. 5. Jing Ze, Fangzheng Xue, Zushu Li. Measurement of spatial target location based on monocular vision [J]. Sensors and Microsystems, 2011, 3(30): 125-128. 6. Lijun Zhang. A multi-layer pipeline parallel processing structure used for offline handwritten Chinese character recognition [J]. Microelectronics, 2000, 30(6): 378 381. 7. Guoxing Li, Bingxue Shi. The design of a digital recognition system for the offline hand of VLSI [J]. Electronic Journals, 1999, 27(11): 143 145. 8. Jiansheng Liu, TongqinWang, Guixin Wang. Based on the layout analysis of DSP and the principle and realization of OCR recognition [J]. Journal of Instrument and Instrument, 2003, 24(6): 577 580. 9. ZhenYin, Quanqi Chen, Yujin Zhang. Deep learning and new progress in goals and behavioral recognition [J]. Journal of Chinese Graphic Graphics, 2014, 19(2): 175-184. 100