Article Preview
Top1. Introduction
DPM is the abbreviation of Direct Part Mark, it takes Data Matrix 2D bar code as the main carrier, Data Matrix bar code has the characteristic of large coding capacity, high density and high information security, etc. Traditional 2D code is basically achieved by printing method, but printing paper products are easy to wear, fuzzy and become unreadable, unable to achieve full life-cycle tracing. One of the biggest features of industrial DPM code is permanence, when the parts are in negative or harsh environmental conditions, direct part mark can achieve permanent and whole life-cycle part tracking ability. DPM coding also has the advantage of uniqueness and “embedded” information inside components, which can prevent the forgery of important industrial products, reduce the risk of copycat (Wang Juan & Wang Ping, 2014). However, due to the diversity generating methods and mental materials, collected industrial DPM bar code images usually exist uneven illumination, low contrast, much noise interference, complex background and other unfavorable conditions. How to identify them quickly and accurately is a problem, and how to locate the bar code area accurately is even more critical and difficult (Gao Tao, Du Xiao-cheng, Fister Jr. Iztok, 2014).
The accurate locating technology of DPM 2D bar code images under complex conditions has always been a hot topic in the bar code study area. At national level, the method based on morphology and contours is widely used. Xie Jun-xi et al. proposed the mathematical morphology based 2D bar code recognition technology; Du Xiu-wei et al. put forward a quick response code locating method based on contour feature; Wei Jin-wei et al. proposed a QR code correction and positioning method based on morphology and hough transform. Afterwards, Wang Wei et al. proposed a machine learning based method, they first filter out the background area, then combine the geometry nature of 2D bar code to detect candidate regions, and use clustering growth method to package bar code region (Wang Wei, He Wei-ping, Lei Lei & Lin Qing-song, 2013); Chen wen-yi et al. put forward an extraction method based on connected region, corner features and edge features (Chen Wen-yi & Chen Bei-min, 2014). In the field of foreign research, Chang C. Alecden et al. apply neural network knowledge to 2D bar code locating technology (Chang C.Alec, Lo Chih-Chung & Hsieh Kuang-Han, 1997), Arnould S. et al. use mathematical morphology method to identify the bar code area (Arnould S., Awcock G. & Thomas R., 1995), Okolnishnikova L.V. et al. use regression iterative approach to locate bar code area (Okolnishnikova L. V., 2001). However, these algorithms cannot get satisfactory results for bar code images under wear, pollution and uneven illumination conditions.