Due to the absorption and scattering effects of the medium, the propagation of light underwater leads to the degradation of images captured for visual monitoring of corrosion defects on the surface of underwater oil casings, which affects the subsequent identification of defect features. To address this issue, an enhanced method for visualizing corrosion defect features on the surface of underwater oil casings was proposed, based on an improved dark channel prior model. First, to address the problems of uneven illumination and overexposure caused by the curved surface of the oil casing, illumination correction was performed based on pixel variations within local regions. Then, to mitigate the weakening of features introduced by homogenization calculations, detail enhancement was achieved through local high-frequency features. Furthermore, the dark channel prior model was applied to eliminate the interference of turbid liquids on the image, thereby enhancing the visualization of corrosion defect features. Experimental results indicate that this method partially restores the color of underwater images, improves image clarity, and enhances visual features. By utilizing YOLOv5 for defect recognition, the recognition rate of images processed with the proposed algorithm is significantly improved.