Abstract: Aiming at the problem that the pores are randomly distributed and have different sizes, which leads to low robustness of pore segmentation, and the noise such as wood fiber, wood ray and axial parenchyma has great influence on the pore segmentation effect, this study proposes an improved K-Means clustering and level-set algorithm for wood cross-section pore segmentation. The improved K-means clustering was used to segment the pore area coarsely, which effectively distinguished the pore area from noise areas such as wood fiber, wood ray and axial parenchyma. Then, the improved level-set algorithm was used for fine segmentation of the coarse segmentation results. The experimental results showed that 98.8% of the pores were segmented accurately and effectively in each microscopic image of wood cross-section, and the segmented pores were basically consistent with the actual pores. In contrast, compared with other algorithms, the improved algorithm proposed in this study improved the average segmentation accuracy of each wood microscopic image by 1.7%. This algorithm can effectively solve the problems of noise influence and randomness of the initial clustering center of traditional K-means clustering algorithm in image segmentation, and has higher robustness and good segmentation performance in the segmentation process of pores with different sizes and random distribution.