The image quality assessment algorithm based on feature similarity (FSIM) has certain limitations to some extent when detecting image step edges. A new algorithm named symmetry feature similarity is put forward by combining the FSIM with the symmetry phase congruency (SPC), which is sensitive to step edges, and calculating the gradient magnitude through the isotropic Sobel operator. The proposed algorithm makes use of the symbols responded to the phase adjacent pixels by SPC and the more accurate weighting coefficient of isotropic Sobel operator to detect and locate the image edge feature. The simulation experiments on various kinds of distorted images are carried out, and the results indicate that the SFSIM is more sensitive to sharper image edge features and Gaussian blurred images, on the basis of retaining the excellent performance of the original method.