Aiming at the problem that the workpiece is difficult for machine vision methods to carry out subsequent dimensional measurements and detect defects on the workpiece because of the reflection problem, a reflective image enhancement algorithm based on improved Retinex (Brightness interval partition Retinex, BIP-Retinex) is proposed. First, the image is segmented by double thresholding to obtain two images divided by brightness interval. Finally, the two images are processed by Retinex algorithm and image fusion is performed to get the resultant image. The experimental results show that the images processed by BIP-Retinex algorithm are tested by the image quality evaluation index, and the peak signal-to-noise ratio (PSNR) value is obtained as 7.539, the structural identifiability (SSIM) value is 0.483, and the root-mean-square error (RMSE) value is 107.056. Compared with other algorithms, BIP-Retinex algorithm has better image quality effect, which improves the problem of overall whiteness and serious distortion in the image processed by the traditional Retinex algorithm.