Abstract: It is really time-consuming for Otsu to seek the optimal threshold for the image composed of multiple targets. Therefore, a simple and efficient histogram-based approach for multi-threshold was proposed. The gray range of the object image was divided into two classes with a threshold value. Following this, the two classes were divided into four classes again by applying the Otsu's bisection method to each of the divided sub-ranges. The experimental results show that the proposed method is suitable for image segmentation, and can be efficiently used for pattern matching and face recognition.

Key words: image segmentation, threshold, Otsu’s method, two-class segmentation, variance

摘要： 大津法对由多个目标组成的图像求最佳阈值时，消耗大量时间。针对此不足,提出简单而快速求多阈值的方法，对已分割的小区间反复利用大津的二分法和方差比值求出需要的阈值。首先，对研究图像的整个灰度值区间利用大津法分成两类；然后，对已分出的两个区间分别应用二分大津法，两类分成四类、八类等。比较实验结果表明，该方法适合图像特征，能快速分割目标图像，可以应用于模式匹配以及人脸识别。

关键词: 图像分割, 阈值, 大津法, 二分法, 方差