Binarization for Document Image Based on Multi-scale Conditional Random Fields
Liu Kun Lv Xueqiang Wang Tao Shi Shuicai
(Chinese Information Processing Research Center, Beijing Information Science &Technology University, Beijing 100101,China)
(Beijing TRS Information Technology Co.Ltd., Beijing 100101, China)
This paper proposes a new algorithm based on multi-scale conditional random fields. This algorithm treats the binarization as a tagging process, using mCRF to label every pixel in the image, so as to realize the binarization of the full image. MCRF of discriminate model can accommodate any of the non-independent features, which makes full use of information in the image. From the result can see this algorithm is better than common threshold method in effect.
刘坤,吕学强,王涛,施水才. 基于多尺度条件随机场的文本图像二值化*[J]. 现代图书情报技术, 2009, 25(4): 79-81.
Liu Kun,Lv Xueqiang,Wang Tao,Shi Shuicai. Binarization for Document Image Based on Multi-scale Conditional Random Fields. New Technology of Library and Information Service, 2009, 25(4): 79-81.
［1］ 陈丹，张蜂，贺贵明. 一种改进的文本图像二值化算法 ［J］. 计算机工程, 2003, 29(13): 85-86.
［2］ He X, Zemel R, Carreira-perpinan M. Multiscale Conditional Random Fields for Image Labeling ［C］. In: IEEE Conference. Computer Vision and Pattern Recognition, 2004: 695-702.
［3］ Derin H, Elliott H. Modeling and Segmentation of Noisy and Textured Images Using Gibbs Random Fields ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1987, 9:39-55.
［4］ Lafferty J, McCallum A, Pereira F. Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data ［C］. In: ICML, 2001: 282-289.
［5］ Hinton G E. Training Products of Experts by Minimizing Contrastive Divergence ［J］. Neural Comp, 2002, 14(8): 1771-1800.
