Print Digits Recognition System Research And Design

With the quick change of modern society and the rapid development of informatization, we are now living in the digital age. The digit is developing toward replacing our language expression and memorization to utterance and written language. From ID card numbers, driver’s license numbers, phone numbers and medical exam reports, a series of digital information representatives which can show identities, abilities and healthy qualities, all need the combination of Arabic numbers1234567890, which are the current+network+program interpretation. Hence, if we want to make the digital information era develop efficiently, and fulfill the aim of high quality living standard, it is crucial to use the safe, efficient and smart ways or the technologically precise digital recognitions.Many researchers make efforts to apply every kind of new knowledge to the pre-processing, feature extraction, classification, such as neural network learning, mathematical morphology, and so on. The key to the performance of system application and its development bottleneck, still lie in the performance of the core algorithm of digital recognition. Its ultimate goal is to study high-speed identification algorithm of the zero error rate or a low rejection rate.In this paper, through the analysis of the background practical value, research status, development trends, and practical significance of digital identification, the author has designed extraction methods, which based on Gabor transform and improving the Numeral characteristics of the invariant moment, and has designed numeral classifier, which is based on the Adaboost algorithm and neural network. The specific works are as follow:(1) To analyze and prove the research background, significance, and the research status at home and abroad of Printing digital recognition system, and then to determine the research contents of this paper.(2) To analyze and make comparisons among Hu invariant moments and different improved invariant moments algorithms. Through adopting the Hu moments and four improved invariant moments, the author has taken feature extractions to printed digital image, and has designed a neural network classifier. Under the condition of MATLAB, the author has done simulation experiments which show the different improved invariant moment algorithms have higher discrimination than common Hu moments.(3) Through adopting the method of principal component analysis, the author has taken feature extraction to printed digital images, and has designed3-nearest neighbor classifier. The simulation experiment shows the recognition rate of this method can reach to92%, if the pivot elements are appropriately chosen. (4) The Gabor transform is a good time-frequency analysis method, To use the Gabor various scale and direction of the printed digital image extraction frequency domain information, and the introduction of the next sampling operation, the frequency domain characterization of printed digital imagefeature vectors, on this basis, the introduction of PCA transform second dimensionality reduction, and designed the Adaboost classifier, PCA and Adaboost algorithm characteristics, the paper design three schemes, the simulation results show that these three programs are made more good recognition rate.(5) To analyze and summarize this paper, and to look forward to what to do next.