Text Image Binarization Algorithm Based On An Improved Two-dimensional Otsu Threshold Segmentation

With the development of the technology,computers using optical character recognition systems(OCR system)to automate the entry gradually replace traditional manual entry.The optical character recognition system can automatically scan the files,classify the files and extracts the information in the fi es.It is faster than the traditional manual entry and can have a high accuracy.The French car rental website also uses the OCR system to analyse the files uploaded by the users and to compare their information.Base on the existed binarization algorithm,this paper proposes a new algorithm combing the background compensation mechanism and an improved two-dimensional Otsu threshold segmentation algorithm.The algorithm is composed by three parts including background compensation,grey space nonlinear mapping and threshold segmentation based-on two-dimensional Otsu algorithm.In the grey space nonlinear mapping,this paper proposes to use an updated function Sigmoid with the factor of black pixel point's ratio.In the threshold segmentation stage,this paper proposes to improve and use an new two-dimensional Otsu algorithm using Gaussian distribution and the simulated annealing algorithm which uses the greedy and stochastic mechanism to reduce the time for finding the approximation of the optimal solution.After the tests on the French ID cards,the algorithm proposed by this paper has a better result on the noise reduction and binarization.With the optimization of the simulated annealing algorithm,the time for the calculation of two-dimensional Otsu algorithm is reduced by 40%.The new algorithm has more practicality in the projects.