Image segmentation is the focus of mobile phone film defect detection. However, the low contrast of captured images often makes image segmentation difficult. In this regard, this paper proposed an improved Retinex enhancement method. The method used Gaussian convolution to estimate the illumination component of defect image to obtain the reflection component, performed adaptive nonlinear transformation on the reflection component, employed contrast-limited adaptive histogram equalization (CLAHE) correction to improve the contrast, used the top-hat transform to eliminate the influence of lighting background, and enhanced the defect image of the mobile phone film. Then, aiming at the incomplete segmentation of the dark details of the defect edge by Otsu's algorithm, a gradient image of the enhanced image was introduced to achieve effective segmentation of mobile phone film defects images. The experimental results show that in the case of low contrast, compared with the original defect image, the image processed by this algorithm has an improved information entropy of about 20%, a contrast of about 100%, and an excellent segmentation effect.