基于灰度直方图和谱聚类的文本图像二值化方法

在自动文本提取中，经定位获得的字符区域需二值化后方能有效识别，由于背景的复杂，常用的阈值化方法不能有效分割自然环境下的字符图像。该文提出了一种基于谱聚类的图像二值化方法，该方法利用规范化切痕(Normalized cut, Ncut)作为谱聚类测度，结合灰度直方图计算相似性矩阵，并通过实验确定最佳的直方图等级数，与通常基于像素级相似矩阵相比，算法的空间复杂度和计算复杂性都大为降低。实验结果表明，针对自然场景下的字符图像，该文方法的二值化结果优于常用的阈值分割结果。

Method of Text Image Binarization Processing Using Histogram and Spectral Clustering

The located text regions need to be binarized for accurate recognition in automatic textural extraction. Due to the complex backgrounds, traditional thresholding methods can not segment the character image effectively from natural scenes. A novel approach of binarization is proposed for gray images. The proposed algorithm uses the Normalized graph cut(Ncut) as the measure for spectral clustering, and the weighted matrices used in evaluating the graph cuts are based on the gray levels of an image, rather than the commonly used image pixels. Thus, the proposed algorithm requires much smaller spatial costs and much lower computation complexity. Experiments on text images in natural scene show the superior performance of the proposed method compared to the typical thresholding algorithms.