融合背景估计与U-Net的文档图像二值化算法 postprint

Abstract: Degraded document images have various degradation factors, such as page stains, ink bleed-through, and background texture. We propose a novel document image binarization algorithm based on background estimation and U-Net. The algorithm first performs image contrast enhancement, and estimates the document background via morphological closing operations. We then adopt a fully convolutional network, namely the U-Net, to extract the foreground text from the document background. Finally, the global optimal thresholding method is used to obtain the resulting binary image. The proposed technique has been extensively evaluated over the recent DIBCO benchmark datasets. Experimental results show that our proposed method outperforms other state-of-the-art document image binarization algorithms in terms of F-measure, pseudo F-measure, PSNR and DRD, with 5.58%、2.47%、0.86dB and 1.19%.