Due to the proximity of the optical diffracting limitation, the edges of the stripes in micrometer scale are blurred after magnification by the optical microscope. Together with the uneven illumination caused by coaxial lighting source, the stripe image quality is very poor. To effectively measure the stripe distance between two stripes in micrometer scale, a centerline extraction algorithm was proposed. First, a single scale Retinex model was adopted to enhance the original image to overcome inaccurate segmentation caused by uneven illumination. After that, binary segmentation was conducted on the enhanced image via optimal threshold value of Ostu. Then, directing at the phenomenon that there was much rag and hollow at the boundary of stripes after segmentation, a boundary collapse method based on fast marching algorithm was utilized to extract centerline accurately. Finally, least square fitting method was conducted to fit the extracted centerlines. Experimental results show that the proposed method can effectively realize accurate extraction of stripe centerline in microscopic image with micrometer scale; the maximum measured deviation of this method is less than 2% when it is used for measuring stripe distance between two stripes with standard width.