Fiber optic endoscope is widely used to detect damage in small space of complex structures, but due to its inherent characteristics and external noise, the acquired images show a honeycomb grid, and there is about 30% barrel distortion. For this problem, a calibration method based on optical fiber endoscope image is proposed. Firstly, a freely moving calibration system is designed, and then the optical fiber bundle in the calibration plate image is processed by Gaussian filtering, binarization processing, mark point identifying and distortion correcting. Finally, the calibration method is used to calibrate and correct different types of calibration plate images taken by optical fiber endoscope. The experimental results show that the proposed calibration method is suitable for correcting and identifying of various scenes and images of different quality, and the measurement error is only 10 μm.