Abstract: In the field of seismic imaging and hydrocarbon exploration, full-azimuth angle-domain common image gathers are of great importance, as they can effectively characterize and represent the propagation behavior of seismic waves in subsurface media and their reflection features. Through the analysis of residual curvature in angle-domain common image gathers, high-precision tomographic inversion can be performed to further improve imaging accuracy. Optimized methods for extracting full-azimuth angle-domain common image gathers were studied. The traditional Horn–Schunck (HS) optical flow method is prone to noise interference from various directions when calculating wavefield directions, and its computational accuracy and noise resistance still need improvement. In this study, the HS optical flow method was improved by developing a modified optical flow approach based on a fractional-order objective function. This method assigned direction-specific weights to both the data term and the regularization term, thereby further constraining the wavefield direction and enhancing computational accuracy and noise resistance. Finally, tests on complex models and real datasets demonstrate that the proposed method can extract higher-quality full-azimuth angle-domain common image gathers in the presence of complex structures and low signal-to-noise ratio field data, with significantly improved wavefield direction estimation and noise resistance, validating the effectiveness of the proposed approach.