直接从双峰直方图确定二值化阈值

DIRECT DETERMINATION OF THRESHOLD FROM BIMODAL HISTOGRAM

图像分割是计算机视觉中重要和基础的研究内容之一,使用阈值是一种常见且重要的区域分割技术,它对物体与背景有较强对比的景物的分割特别有用.通常采用直方图技术来确定阈值.本文介绍了一种新的不需对直方图做预处理而能直接通过计算求得二值化阈值的方法,并将它与常用的大津法(Otsu法)进行了比较.实验表明,本文的方法是快速有效的.

In this paper, a new and fast method of determining the binarizing threshold for a image with high contrast directly from its bimodal histogram is proposed. Instead of preprocessing on histogram such as smoothing or curve fitting to eliminate the effect of noises, statistic characteristics of the histogram is utilized to coarsely estimate the ranges of gray values of objects and background. Then the most frequent value in each range, which can represent the typical pixel values in objects or background, is found as the peak position. The threshold is set at the middle of the two peak positions. Experiments show that the result is comparable with Otsu method. Because of small time consuming and the capability of anti-noise, the proposed method is suitable for applications with strict speed requirement.