A new binarization method for high accuracy handwritten digit recognition of slabs in steel companies

摘要

Handwritten digit recognition plays a crucial role in various automatic systems and industries nowadays. In steel companies, the code digits written on slab forehead should be recognized with high accuracy because these codes are used for determination of suitable thickness of steel slabs in rolling procedure. Because of special artifacts exist in these kinds of images such as color sagging, faint colors, and color separation, most of the common digit recognition algorithms have low accuracy in recognition of code digits. In this paper, a new two stage algorithm is proposed in order to reach high accuracy in determination of the handwritten codes written on slabs forehead and are corrupted by color sagging, faint colors and color separation. Applying our proposed algorithm on a dataset of images of slabs in Isfahan Steel Company shows that accuracy of the algorithm is 89.12% which is much higher than common and recently proposed algorithms.