The Research On Character Recognition Technique In The Fieldwork Of Industry

The character recognition technique is one of the most important research directions in the field of pattern recognition by machine vision. With the improvement of automatic level in modern production, many character signs of products need to be recognized on- line by machine automatically. In the fieldwork of industry, many disadvantages, such as abominable environment, unbalancing illumination, displacement of carried products, etc.. will cause that the obtained images are fuzzy, noisy or will bring characters into scaling, shift and rotation. What the author has studied and realized by Matlab programme mainly are as follows: (1) The pre-processing for image: (2) Presenting a novel binarization method based on image partition derived from Da-Jing. Experiments has demonstrated that the fuzzy image with unbalancing illumination could be binarized perfectly by this method. (3) Feature extraction. This paper mainly studied the moment features which are invariant to rotation in theory. After normalizing to characters we selected 18 Zernike moments as feature finally. Otherwise the gridding feature was also extracted; (4) Designing three neural network (NN) recognition systems using the modified BP learning algorithm for two features; (5) Using three systems to recognize the characters from the trainwheel. The experiments demonstrated that the gridding feature network has a good ability of robust for salt & pepper noise; the Zernike moments network can be employed successfully for rotation-invariant characters. Combining two features for two-level NN classifiers to recognize character can obtain the correct recognition rate of 98%.This project is supported by the Science & Technology Research Foundation of...