The Recognition Technology Research Of Low Quality DM Code Image

For the moment, the Internet of things develops rapidly and its application cannot do without automatic recognition. Two-dimensional code, as a new information transmission media, can meet the strong demands due to its big information storage, accuracy and real-timing. In consequence, many domestic companies and research organizations invest substantial resources in technology research and development.Currently, the standard of two-dimensional encoding has become general. Therefore, the main problem is how to decode two-dimensional image using the existing standard. However, the captured code images are often in low quality according to shooting angle, illumination and printing device. This thesis is focusing on how to reconstruct code structure from the low-quality DM images, and the main research contents are as follows:1. Two-dimensional barcode image preprocessing:image enhancement processing is used to improve the contrast of barcode images;2. Improved L shaped positioning algorithm:for some images, L shape feature is so fuzzy that we cannot accurately locate the barcode position. We proposed an algorithm by combining Max-Min algorithm and Canny edge detection algorithm. It can greatly improve the L shape feature.3. Adaptive structure localization method of local feature:Irregular deformation may lead to the uneven distribution of the barcode module. Moreover, distortion may lead to recover barcode more difficultly. We proposed an adaptive adjustment method to using the edge features of the module. In this way, the division results can reconstruct the original structure more accurately. Experiments showed the proposed algorithm can get better results over the existing methods.4. The implementation of the core components of two-dimensional barcode recognition:Base on the above research, the thesis developed a core component of two-dimensional barcode recognition, and applied it to information collection of the social security and industrial code recognition. The system consists of image pre-processing, structure recovery and barcode decoding modules. We tested the proposed method by lots of real data. Results show our method is stable and robust.