Research on online recognition system of finished tobacco box number and key technology

For the low efficiency of online detection of the finished tobacco box number (FTBN), an online detection system based on machine vision was introduced. To avoid the edge interference caused by non-uniform illumination, region projection technology was used to achieve the vertical position of the FTBN. A slant correction approach based on the symmetry centers of coarse grids was proposed, and efficiently solved slant correction problem of the number without trail lines. Then a segmentation method of two-step projection was introduced to segment characters precisely. In the number recognition stage, the statistic features of coarse grids were extracted and the FTBN was recognized by improved BP algorithm. The final test results and field application shows that the algorithm is efficiency and the system is reliability.