Using Convolutional Encoder-Decoder for Document Image Binarization

摘要: Document image binarization is one of the critical initial steps for document analysis and understanding. Previous work mostly focused on exploiting hand-crafted features to build statistical models distinguishing text from background. However, these approaches only achieved limited success because: (a) effectiveness by researcher's domain knowledge understanding documents, (b) a universal model cannot always capture complexity different degradations. In order address challenges, we propose convolutional encoder-decoder with deep learning in this paper. proposed method, mid-level representations are learnt stack layers, which compose encoder architecture. Then obtained mapping low resolution original size through decoder, composed series transposed layers. We compare method other algorithms both qualitatively quantitatively public dataset. The experimental results show that has comparable performance more generalization capabilities in-domain training data.