Abstract: The identification of fractures in the surrounding rock of the roadway is an important step in achieving the perception of the structural characteristics of the rock mass to be excavated underground. Currently, the quantitative influence mechanism of the unevenness of the rock surface on the accuracy of fracture identification has not been fully explored. This paper used the joint roughness coefficient (JRC) as a quantitative indicator of the unevenness of the rock surface, combined the Weierstrass-Mandelbrot fractal function with high-precision photopolymer 3D printing technology, and realized the preparation of fracture-containing rock test samples covering all levels of JRC. In a controlled experimental environment, sample images were collected, and the images were preprocessed in a standardized manner through grayscale conversion, local histogram equalization, and adaptive gamma correction. The performance of three algorithms, namely Canny edge detection, Otsu threshold segmentation, and region growing, in fracture identification under different JRC conditions was systematically compared. The following conclusions were obtained: The influence of rock surface unevenness on the accuracy of fracture identification intensifies with the increase of JRC, and the responses vary due to different algorithm mechanisms: The Canny algorithm is interfered by the "texture gradient confusion" mechanism, and the error rate is always higher than 18% within the full JRC range, and the recognition rate drops sharply after JRC > 10; The Otsu algorithm is interfered by the "local gray level distortion" mechanism, and the double-peak structure of the gray level histogram is destroyed, and the recognition rate drops to below 83% after JRC ≥ 10; The region growing algorithm effectively avoids the interference of the above two mechanisms by relying on the dual constraints of local gray similarity and spatial connectivity, maintaining an identification rate of over 98% and an error rate of less than 7% within the full JRC range, demonstrating strong robustness. The study reveals the intrinsic influence mechanism of rock surface unevenness on fracture image recognition, providing experimental basis and algorithm reference for the development of visual recognition systems under complex and rough rock surfaces in underground conditions.