Research Of Off-line Handwritten Character Recognition Algorithm For Specific Character Set

Security clearance is an important means to protect national information Security. With the raised strength of security clearance, the clearance of handwritten documents becomes a main object of research applied to the tools of security clearance. Because of unconstrained strokes and uncertain stroke orders, off-line handwriting Chinese recognition is always a big challenge. And it is far below the requirements of Security clearance when aiming at special characters like security, national and safety. This paper aims to find an effective way to realize the function of off-line handwriting Chinese recognition including noise suppression,stroke recovery and feature recognition.This paper focuses on several key steps of handwriting recognition and the main work are as below:Firstly, it improved preconditioning from two aspects: removing noise taken by thinning and recovery of stokes in unconstrained handwritten Chinese. To solve the problem of noise taken by thinning,we describe a way to determine the nodes of stokes to distinguish crossing points, end points and inside points and then we can recovery this wrong stokes. To solve the problem of unconstrained handwriting, we treat computers with the help of artificial neural nets to remove the otiose stokes and supplement the missing stokes. This can greatly reduce the difficult of off-line handwritten Chinese recognition caused by different handwriting habit.Secondly,it improved the step of feature extraction,designed a classifier and proposed a algorithm of feature matching based on the classifier. It described the advantages and disadvantages of two ways of feature extraction and proposed a restoration method of actual feature of the stokes. After extracting, we classify the features roughly and design a classifier with three layers to avoid waste of time when lexicographic search. We established a database of three-layer strokes feature library and put forward the matching method and the matching principle to calculate similarity between Chinese to be recognized and the strokes feature library.Thirdly, combining with the improvement proposed by this paper and the requirements of the tools of security clearance, we designed and developed a set of document recognition system and analyzed the data acquired. The experiment on 1530 samples of 6 Chinese characters extracted from a specific character shows that while the recognition rate of mainstream recognition software is not exceed 50%, this document recognition system can still maintain the recognition rate 90% at least.Experiments show that the recognition rate of specific character can be improved significantly. If make more template, the recognition rate will continue to rise. So in the same time and the same environment,this system is the most suitable for the test tools of security clearance.