Toward a Binarization Framework resolving the Maghrebian Font Database challenges

摘要

Historical documents are of major importance in conserving cultural and scientific heritage. The Maghrebian font of the Arabic language had dominated in different countries. Thousands of cultural and scientific documents were written using this style. In this paper, we intend to present the built Maghrebian font database with giving the different challenges facing the development of an Optical Character Recognition system able to treat it. Also, we are going to reveal the new developed binarization framework, based on Deep Learning theory, able to get the best accuracy compared to different studied binarization approaches. The proposed framework is based on Selectional Auto-Encoders and the Energy minimization theory. The obtained binarization result on the Maghrebian font using the F -measure is 98,1%. In this paper, we intend to present the built Maghrebian font database with giving the different challenges facing the development of an Optical Character Recognition system able to treat it. Also, we are going to reveal the new developed binarization framework, based on Deep Learning theory, able to get the best accuracy compared to different studied binarization approaches. The proposed framework is based on Selectional Auto-Encoders and the Energy minimization theory. The obtained binarization result on the Maghrebian font using the F -measure is 98,1%. Also, we are going to reveal the new developed binarization framework, based on Deep Learning theory, able to get the best accuracy compared to different studied binarization approaches. The proposed framework is based on Selectional Auto-Encoders and the Energy minimization theory. The obtained binarization result on the Maghrebian font using the F -measure is 98,1%.