Batch Number Of Billets Recognition System Design

Character recognition is one of the most important application in computer vision. Character recognition play a great role in various OCR systems, LPR systems and so on. Batch number of billets recognition system is yet another application of character recognition. Batch number of billets recognition refers to automatic acquisition of billet images and recognition of batch number using related techniques in computer vision. To increase the degree of automation and reduce the labor intensity of worker in rolling mills, it is necessary to establish bath number of billets recognition system. Batches of steel billet are used to indentify furnace, production date and so on. The batch number can be used to track the billet in the whole process of production and transportation.This article was based on the application described above. Some similar applications are noticed in the computer vision. This article first compared the automatic license plate recognition system and batch number of billets recognition system, and then gave an overall framework of batches recognition system. Three techniques are considered important in the batches number recognition system: binarization, character segmentation and character recognition.This article first discussed various global threshold binarization algorithms and local adaptive threshold binarization algorithms, and then a combination of Otsu global threshold binarization and Niblack local adaptive threshold binarization was proposed. Then this article discussed two commonly used character segmentation algorithm: projection analysis and connected components analysis. Based on research of those two segmentation algorithms, a new segmentation algorithm was proposed, which combined projection analysis and connected components analysis together to get a better result. In the following chapter, this article discussed two widely used character recognition algorithm: template match and neural network. Then this article focused on how to choose the best feature to describe those characters to be recognized. At last, projection feature was chosen as the global feature and Zone feature as the local feature. A three layer BP artificial neural network was trained to recognize characters described by those two kind of features. In the last part of this article, test result and some test result analysis were presented to demonstrate the algorithm proposed above, which proved to have practical utility.