Markov random field based binarization for hand-held devices captured document images

In this paper, a novel Markov random fields (MRF) based binarization algorithm is proposed to segment foreground text from document images captured using hand-held devices (such as cell-phone or digital camera). In the MRF based framework, an edge potential feature is extracted to preserve the strokes of foreground text and to remove isolated noise and an intensity feature is used to smooth the entire document image. Prior to binarization, we use a nonlinear function to enhance the quality of document images which suffer from insufficient or uneven illumination. Experimental results show that our method outperforms other state-of-the-art approaches.