CT scan test during the destruction of coal rock specimens can realize the visual characterization of the internal fracture structure of coal rock specimens, but during the loading process, the displacement and deformation of the specimens can cause inaccurate positioning of the scanned layers, which in turn affects the test results. To solve this problem, the author used image similarity retrieval technique to retrieve the most similar CT images of the loaded specimens at different stress levels, so as to more accurately describe the fracture evolution law of the loaded coal samples. The results show that (1) during triaxial compression damage, the specimen stress-strain curve was divided into five distinct phases: the initial compression-density phase, linear elastic deformation phase, plastic yielding phase, peak damage phase, and residual deformation phase. (2) The image retrieval technique based on perceptual hash algorithm could accurately identify and locate the CT images with the highest similarity at different stress stages, and the similarity of CT images gradually decreased with increasing axial load. (3) During the whole triaxial loading process, the mean value of the absolute difference of similarity of CT images exhibited two obvious stages of change: the slowly increasing stage and the rapidly increasing stage. (4) The fractal dimension can quantitatively describe the fracture evolution, and the fractal dimension mainly went through three stages of slow decrease, slow increase and rapid increase during the damage of loaded samples.