When eddy-current infrared thermal-imaging technology is used to detect metal-material damage defects, the infrared image is susceptible to noise and may also contain useless information, which can result in blurring of damage defects. To address this problem, a defect-detection method based on single-scale Retinex and improved K-means clustering is proposed to perform infrared image-feature enhancement, image segmentation, and edge feature extraction. First, the image is enhanced using single-scale Retinex. Additionally, the defect features are enhanced. Then, an improved K-means clustering algorithm is used to segment the image. Finally, a mathematical morphology algorithm is used to process the image, which removes the useless information in the defective image and uses a Canny operator to detect the defect edge. The experimental results show that the method effectively detects defects of metal-material specimens and extracts complete and clear defect edges of the metal-material specimens.