| This thesis analyzes the process of tobacco packaging cause cigarette packet printing character recognition with low recognition rate for several reasons, given the tobacco manufacturing industry field conditions, will cause distortion, fuzzy, noise and image of the characters of the scaling, translation, rotation, and so on and so forth, a targeted cigarette packet printing characters based on BP neural network, intelligent recognition system has carried out research.Firstly, the paper introduces in detail in the process of cigarette packet printing character recognition using some of the basic technology, including image preprocessing, feature extraction, image segmentation and image recognition, the key link. The paper was designed a set of perfect preprocessing scheme user the histogram of image is described, and based on this design the related such as edge detection, image enhancement, noise removal algorithm, which solve the cigarette packet printing all kinds of character images in the process of character recognition problem. In printing character feature extraction phase, by analyzing characteristics of unicom characters printing, and the histogram of the vertical projection for character segmentation caused by the adverse results, is proposed according to the code character of upper and lower boundary of the projection method, this method of printing characters segmentation is more reliable, improve the printing character recognition rate.Then, the thesis was introduced the neural network, points out that the artificial neural network is a kind of computer theory, mimicking the biological brain function, especially BP neural network has a good practical results for solving image recognition problems, on this basis, the author proposing the introduction of steepness factor method to improve the ability of artificial neural networks for character recognition, through the experiment, the improved algorithm can obviously improve the character recognition character recognition rate of the system, is a very effective method of character recognition. |
