Rock fractures have a strong impact on the radionuclide transport and retention. We sampled a number of rock specimens and then acquired the rock fracture images. The quaternion convolution can be used to detect the rock fracture edges. In this paper, a simplified quaternion convolution algorithm is presented and a set of quaternion masks at different scales are presented. The signs are computed for modulus maximum suppression in the horizontal and vertical direction respectively. In order to suppress the noise, we use the average weighting for the results at different scales. Experimental results show that the proposed method is effective and robust. Moreover, this method has an important contribution to rock engineering.