中文摘要：

目的通过白细胞分类计数的自动化获得一种有效的图像描述量，进而实现自动化识别的关键问题。方法本文首先用改进的变形雅可比（P=4，q=2）-傅里叶矩对白细胞显微图像进行尺度、灰度、旋转、平移等多畸变归一化，使显微图像的归一化理论得以补充。然后收集7种白细胞的134种显微图像做训练样本集，用最小平均距离规则对部分白细胞变形图像进行初步识别实验。结果该不变矩具有很强的显微特征提取能力和抗畸变性能。结论白细胞分类计数能有效实现图像的自动化识别。

英文摘要：

Automatic classification and counting of leucocytes is one of the most important research topics in clinical medicine, and how to get an effective image descriptor is the crucial problem for realizing the automatic i- dentification. The microscopic characteristics of leucocytes were theoretically normalized to scale, intensity, rotation and shift multi-distortion invariant according to improved Pseudo-Jaeobi （p =4, q =2）-Fourier Moments in this pa- per, and then 70 variant images of 7 kinds of leucocytes were classified by using the weighted minimum-mean-dis- tance rule. The results showed that the moments not only have an ideal ability of feature extraction, but also have strong properties of anti-multidistortion.