Abstract: Machine vision has become an important approach for on-machine measurement of tool wear. In small-batch machining of key components in fields such as aerospace, cutting experiments on difficult-to-machine materials such as titanium alloys are costly, and the cutting process is irreversible. Therefore, degraded samples are difficult to acquire again. In this study, 104 flank face images of cutting tools were collected from the same experiment. Among them, 22 images were degraded because of machine tool micro-vibration, local defocus, and light reflection. To address this problem, an on-machine measurement method for tool wear based on degraded image enhancement is proposed. Real-ESRGAN is used to restore the degraded images. Then, gray-level enhancement, filtering, threshold segmentation, and morphological processing are performed. The Canny operator is used to extract the initial edge. Zernike moments are then used for subpixel correction. After outlier removal and curve smoothing, a continuous wear boundary is obtained. A side-edge reference line is established to calculate the maximum wear width VB_max. The results show that, for 10 typical degraded samples, the mean absolute error decreases from 63.48 μm to 12.95 μm, and the mean relative error decreases from 25.63% to 5.19%. In comparative experiments, the average absolute error and relative error for the 22 degraded images were 14.49 μm and 6.73%, respectively. For all 104 images, the mean relative error is 4.86%. The proposed method effectively improves wear boundary extraction from degraded images and enhances the stability of VB_max measurement under complex on-machine imaging conditions.