Due to the complexity of cell structure and the overlap of cells, accurate segmentation of cytoplasm and nucleus remains a challenging problem. Here an algorithm based on feature weight adaptive K-means clustering to extract complex leukocytes is proposed. In traditional Kmeans clustering algorithm, the initial clustering center is randomly assigned, which affects the clustering effect. In this paper, the initial clustering center is selected according to the histogram distribution of cell image, which not only improves the clustering effect, but also reduces the time complexity of the algorithm from O (n) to O (1). Then, the improved K-means algorithm can have some anti-noise performance by using a non-Euclidean distance．Before leukocytes are extracted, the color space is decomposed, and the cell nucleus and cytoplasm are extracted according to the color component and the improved K-means clustering algorithm. Color space decomposition and K-means clustering are combined for segmentation. Finally, Adherent leukocytes are separated based on watershed algorithm. The proposed segmentation method achieves 95.81% and 91.28% overall accuracy for nucleus and cytoplasm segmentation, respectively. Experimental results show that the new method can effectively segment complex leukocytes and have high accuracy.