To meet the requirements of rapid detection of fiber cross-section geometry parameters in the measurement of industry, digital image processing technology is used to realize fiber geometric parameter measurement. The high-contrast fiber secondary coating layer region is segmented based on Otsu method; the edge detection is performed on the low-contrast fiber primary coating layer and cladding layer based on the improved Canny algorithm. The algorithm uses the local region distance threshold based on the traditional Canny algorithm. The method replaces the high threshold in the double threshold method to realize the single edge response principle; the curve is fitted to the edge based on the combination of the quadratic curve normalization method and the homogeneous equation method, and the Gauss-Newton iteration method is used to solve the fiber geometric parameters. The test results are compared with the high-precision fiber tester FGM-502, the results prove the validity and accuracy of the fiber detection algorithm studied in this paper, and the accuracy meets the requirements of enterprise fiber geometry parameter detection.