Abstract: The new image segmentation algorithm based on generalized harmonic mean distance is proposed.The Otsu’s thresholding method is a kind of representative image thresholding segmentation algorithm，and it has widely applied in image segmentation field by reason of its many characteristics such as computing simplicity，fast speed and performance stability.The traditional Otsu’s thresholding method is a kind of minimum variation thresholding algorithm based on Euclidean distance and is devoid of robust performance in image segmentation because of Euclidean distance without adjustable parameter.This paper first replaces the Euclidean distance in Otsu’s thresholding segmentation method with the generalized harmonic mean distance and obtains a kind of new robust image segmentation algorithm，then the method of choosing its optimal parameter in new segmentation algorithm is brought forward.A large quantity of experimental results show that the new segmentation algorithm is more effective than otsu’s method.

摘要： 提出了基于广义调和均值距离的最小偏差图像阈值化分割新算法。Otsu阈值法是图像分割中最典型阈值法之一，因其计算简单、速度快和性能稳定等优点而在图像分割中得到广泛应用；但是，传统Otsu阈值法是基于欧式距离的最小偏差阈值法，由于欧式距离没有可调节参数而导致Otsu阈值法分割图像缺乏鲁棒性。首先将Otsu图像分割法中的欧式距离用广义调和均值距离代替并得到一种具有鲁棒性的图像分割新算法，其次给出该算法中参数选取办法。大量实验结果表明，新的图像分割算法相比Otsu法更有效。