Recognition results classification and post-processing methods for painted characters on billet surface

Abstract

Automatic identification of characters marked on billets is very important for steelworks to achieve manufacturing and logistics informatization management. Due to the presence of adhesions, fractures, blurs, and other problems in characters painted on billets, character recognition accuracy with machine vision is relatively low, and hardly meets practical application requirements. To make the character recognition results more reliable and accurate, an identification results classification and post-processing method has been proposed in this paper. By analyzing issues in the image segmentation and recognition stage, the recognition result classification model, based on character encoding rules and recognition confidence, is built, and the character recognition results can be classified as correct, suspect, or wrong. In the post-processing stage, a human-machine-cooperation mechanism with a post-processing interface is designed to eliminate error information in suspect and wrong types. The system was developed and experiments conducted with images acquired in an iron and steel factory. The results show the character recognition accuracy to be approximately 89% using the character recognizer. However, this result cannot be directly applied in information management systems. With the proposed post-processing method, a human worker will query the suspect and wrong results classified by the system, determine whether the result is correct or wrong, and then, correct the wrong result through the post-processing interface. Using this method, the character recognition accuracy ultimately improves to 99.4%. Thus, the results will be more reliable applied in a practical system.

Keywords

Painted character / Character segmentation / Character recognition / Recognition results classification / Post-processing method

References

[1] Alexandria AR, Cortez PC, Felix JHS, et al. An OCR system for numerals applied to energy meters. IEEE Latin Am Trans, 2014, 12(6): 957-964.

[2] Deng CG, Xiang ZQ, Xie JC, et al. Design and study on marking system online for steel products. Comput Integr Manuf Syst, 2010, 16(9): 1859-1865.

[3] Agarwal K, Shivpuri R. Knowledge discovery in steel bar rolling mills using scheduling data and automated inspection. J Intell Manuf, 2014, 25(6): 1289-1299.

[4] Jia HB, Murphey YL, Shi JJ (2004) An intelligent real-time vision system for surface defect detection. In: Proceedings of the 17th international conference on pattern recognition, Cambridge, UK, pp 239–242

[5] Zhu JN, Mae Y, Minami M. Finding and quantitative evaluation of minute flaws on metal surface using hairline. IEEE Trans Ind Electron, 2007, 54(3): 1420-1429.

[6] Bulnes FG, Usamentiaga R, Garcia DF (2011) Vision-based technique for periodical defect detection in hot steel strips. In: Proceedings of IEEE international conference on industry applications society annual meeting, Orlando, FL, pp 1–8

[7] Choi SH, Yun JP, Sim SB (2010) Edge-based text localization and character segmentation algorithms for automatic slab information recognition. In: Proceedings of IEEE international conference on image analysis and signal processing, Zhejiang, China, pp 387–392

[8] Bulnes FG, Usamentiaga R, Garcia DF (2011) Detection of periodical patterns in the defects identified by computer vision systems. In: Proceedings of IEEE international conference on intelligent systems design and applications, Cordoba, Spain, pp 301–306

[9] Li P, Wu SB, Wang LN, et al. Automatic recognition system of plate character based on machine vision. Metall Ind Autom, 2013, 37(4): 12-15.

[10] Novak V, Hurtik P, Habiballa H, et al. Recognition of damaged letters based on mathematical fuzzy logic analysis. J Appl Logic, 2014, 189(2): 497-506.

[11] Mi ZZ, Xie ZJ, Liu Q. Study of heavy rail mark recognition method based on PSO-SVM. Mod Sci Instrum, 2016, 6: 87-90.

[12] Zhao QJ, Cao P, Tu DW. Toward intelligent manufacturing: label characters marking and recognition method for steel products with machine vision. Adv Manuf, 2014, 2(1): 3-12.

[13] Huang GB, Kae A, Doersch C, et al. Bounding the probability of error for high precision optical character recognition. J Mach Learn Res, 2012, 13(1): 363-387.

[14] Rico-Juan JR, Calvo-Zaragoza J. Improving classification using a confidence matrix based on weak classifiers applied to OCR. Neurocomputing, 2014, 151: 1354-1361.

[15] Navarro-Cerdan JR, Arlandis J, Llobret R, et al. Batch-adaptive rejection threshold estimation with application to OCR post-processing. Expert Syst Appl, 2015, 42(21): 8111-8122.

[16] Qiu Q, Li Z, Ahmed K, et al. A neuromorphic architecture for context aware text image recognition. J Signal Process Syst, 2016, 84(3): 1-15.

[17] Zhu A, Gao R, Uchida S. Could scene context be beneficial for scene text detection?. Pattern Recognit, 2016, 58: 204-215.

[18] Quehl B, Yang H, Sack H (2015) Improving text recognition by distinguishing scene and overlay text. In: Proceedings of SPIE the 7th international conference on machine vision. doi:10.1117/12.2181370

[19] Serrano N, Civera J, Sanchis A, et al. Effective balancing error and user effort in interactive handwriting recognition. Pattern Recognit Lett, 2013, 37(1): 135-142.

[20] Betta G, Capriglione D, Corvino M, et al. A proposal for the management of the measurement uncertainty in classification and recognition problems. IEEE Trans Instrum Meas, 2015, 64(2): 392-402.

[21] Kim S, Yu Z, Kil RM, et al. Deep learning of support vector machines with class probability output networks. Neural Netw, 2015, 64: 19-28.