Character Recognition Research Based On Artificial Intelligence And Machine Learning

In recent years, machine learning becomes a new research focus in the field of artificial intelligence. It has been successfully applied in the complex systems such as: machine vision, speech recognition, natural language processing, web search, recommendation system, intelligent robots etc. Especially, in the last two years the appears of the autopilot, deep QA system which based on artificial intelligence and machine learning technology make people began to rethink the word:machine is invented by human, it can never exceed the level of human intelligence。The Chinese character recognition has been a difficult problem in the field of character recognition. Different from the English text consisting of a small number of characters, it is difficult to use traditional algorithm to identify it automatically. But thanks to the further development of machine artificial intelligence, the automatic identification of Chinese characters has entered the practical stage. Although many domestic and foreign software vendors have launched a rate of Chinese characters automatic identification system which has a good recognition, there is still large room for improvement.In a large number of current domestic literatures, mainly papers aim at the research on automatic recognition of a small amount of characters. It is difficult to be applied to large character set recognition object. This is closely related to the structure of machine learning and learning algorithm. Satisficing votes of each classifier, which is trained pre viously to classify the characters feature vector, and taking the result of most votes as th e final output is the current foreign mainstream solution.Using the medical record sheet as the recognize object and In line with the idea of multi-angle recognition and cross-validation, this paper extracts sets of features from the character image, train multiple classifiers. Finally, under certain fault conditions, compare the input feature vector of each classifier and the pre-saved feature vector corresponding to the classifier’s output, output the result which match the pre-saved feature best. Experimental results show that this method not only can correctly identify the inputs, it can also discover the error recognize result itself. This means that it will be available to realize a character recognition system which has self-discovery and self-correction function. In addition, this paper evaluates the performance of support vector machine (SVM) and BP neural network classifier, which has some guidance on the choice of character recognition in learning.