We report a machine vision system with high dynamic range designed for optical surface defect inspection. The system consists of two motorized linear stages and one motorized rotation stage for automatically scanning the surface. As the intensity of long scratches and digs differ a lot under dark-field illumination, gains of red, green, and blue channels are set to be different values to extend the dynamic range of an ordinary colored detector in a single snapshot, which greatly improves the efficiency compared with multiexposure-based approaches. Image stitching is then employed to get a high-resolution image of the entire optical surface for further processing to quantitatively evaluate surface quality based on the standards of MIL-PRF-13830B and ISO 10110-7:2008. The system can be widely applied in the optical industry, as it provides a low-cost solution for optical surface quality checks.