几种文本图像二值化方法的对比分析

Comparison and Analysis of Several Document Image Binarization Algorithms

文档图像二值化是光学字符识别(OCR)预处理阶段的关键技术,其处理效果的好坏直接影响光学字符识别的识别率的高低.本文对几种全局阈值算法与自适应局部阈值算法进行了研究分析,从理论模型和实例文档图像两个方面综合分析了OTSU法、迭代法、简单统计法、Bersen、Niblack、VFCM、NFCM等二值化算法.通过实验、对比分析表明,几种算法各自具有其优缺点,对一些复杂文档图像情况的适应性也各不相同.在实际应用中,应根据具体的文档图像质量、周围环境的光照,以及对图像处理时间的要求等情况选择合适的文档图像二值化算法.

Document image binarization is the key technology of Optical Character Recognition（OCR）＇s preprocessing.The result of the process directly influences OCR system＇s recognition rate.In this paper,several global threshold algorithms and adaptive local threshold algorithms are researched and analyzed.The methods of OTSU,iteration,simple statistics,Bersen,Niblack,VFCM and NFCM are compared from the aspects of theoretical models and case studies.Experimental results show that each algorithm has its own advantages and disadvantages.Under different document image situations,these algorithms also have different adaptabilities.In practical application,the appropriate algorithm for the document image binarization should be selected according to different document image quality,surrounding light environment,and the time demanded of image processing.