针对相差显微镜采集的细胞图像具有亮度不均衡且细胞与背景对比度较低的问题，根据图像中细胞区域灰度变化急剧和背景区域灰度变化缓慢的特征，提出一种基于双重高斯滤波的细胞图像快速分割方法。从频率域的角度出发，构建双重高斯滤波过滤低频信息、保留高频信息，有利于增强细胞与背景的差异性，降低亮度不均衡因素的干扰；借助自适应阈值分割出细胞，并使用形态学闭运算改善其形态，同时利用面积约束在一定程度上消除杂质对分割准确性的影响。在C2C12数据集上测试所提方法，精确率、召回率和F值分别为0.9770、0.9457和0.9609，优于对比算法。研究结果表明：所提方法具有较好的准确性和鲁棒性，能实现细胞图像快速、准确分割。

A problem in terms of the accuracy of noninvasive measurement of blood glucose with near-infrared (NIR) spectroscopy is mainly caused by the weak glucose signal and strong background variations. We report the existence of the radial reference point in a floating-reference method, which is supposed to solve this problem. Based on the analysis of the infinite diffusion theory, the local condition of the reference point is deduced theoretically. Then the experiments using the intralipid solutions are constructed to testify the existence of the reference point. In order to further validate our results, Monte Carlo simulations are performed to calculate the diffused light distribution according to the variation of the glucose concentration in the intralipid solutions. All the reference points existing in three-layer skin model are also listed at the wavelength of 1200-1700 nm.