A simple structure, clear imaging optical microscopic imaging system aiming at the requirement of long working distance and large tilt angle for the cell factories was designed. According to the characteristics of the cell images uneven illumination, sample turbidity, cell adhesion, cell overlap and the unobvious cell distance, the Retinex algorithm was adopt to pretreat the image, then a fast and compositive image segmentation algorithm based on the Otsu and K-means clustering was used to segmented the cell images, at last, an improved fast connected component labeling method and a rapid high precision cell counting method were applied to count the number of cells. The results of simulation experiment and actual test show that the imaging quality of the system achieved the project actual requirements for resolution and definition, and the morphological and distribution of cells in culture dish could be identified clearly. The cell microscopic image processing method has a good image enhancement effect, weakens the phenomenon caused by the uneven illumination and sample turbidity that human vision cannot clearly distinguish the cell, and eliminates the statistical error caused by that image segmentation is not in place. The cell counting accuracy is above 95%. The method is suitable for various cells, which can meet the requirements of real-time and off-line monitoring for cell factories.