Research On Partial Differential Equation Model For Document Image Binarization

Document image binarization(The conversion of the grayscale or color document images into binary images)is the basis of optical character recognition,which plays a vital role in document image analysis and pattern recognition.In recent years,the research on document image binarization has received extensive attention.In particular,the successful launch of the first Document Image Binarization Competitions(DIBCO)in 2009 represented a milestone for the development of document image binarization.The DIBCO series datasets(DIBCO 2009-2014,DIBCO 2016-2018)provides the rich test images for the study of the binarization document image.The quantitative indexes used in the DIBCO competition provide a reliable basis for evaluating the quality of the document image binarization results.Image acquisition process or document itself may cause document image degradation,such as uneven illumination,noise,contrast variation,bleed-through and blur.Therefore,the requirement for binarization techniques for degraded document images is increasing.In recent years,the partial differential equations method for document image binarization has been widely paid attention by scholars at home and abroad.Compared with other binarization methods(such as clustering method and threshold method),this method has strong local adaptability and high flexibility.This paper proposes a partial differential equation model for document image binarization and its numerical algorithm.The basic idea of the model is that the original document image is evolved under the control of partial differential equations,resulting in a series of images that tends to be binarized.When the evolution of partial differential equations reaches the steady state,the steady state solution is the desired binarization result.For the numerical implementation of the model,a three-step splitting scheme is designed,which is to decompose the evolutionary partial differential equation into two linear differential equations and one nonlinear differential equation.In each iteration process,the above three equations are solved sequentially by the finite difference method.Model and algorithm performance tests are tested on DIBCO datasets,and F-Measure,Pseudo-FMeasure,PSNR and DRD are adopted as quantitative indexes,in comparison with four partial differential equation models and a classical threshold method(Otsu method)in the literature.The experimental results show that the proposed model has achieved the best results on average for degraded document images such as uneven illumination,contrast variation,blur and bleed-through.The main contribution of this research is to provide a good alternative method for document image binarization technology and enrich the application of partial differential equations in the field of document image binarization.