In the production process of fiber image inverters, it is easy to produce chicken defect, spot and other defects, and how to automatically identify the defects has been the focus of research. A threshold segmentation method is proposed which smoothes the obtained filter image to get an average brightness image. And the average image is subtracted from the original image to get a difference image. The difference image is binarized by a threshold to realize the automatic segmentation of the defect. In the experiment, the effectiveness of the above algorithm is verified by using the typical image of the chicken and the spot defects, which indicates that the method can meet the requirements of fiber image inverters.