Characterization of seismic geologic structures, such as describing fluvial channels and geologic faults, is significant for seismic reservoir prediction. The coherence algorithm is one of the widely used techniques for describing discontinuous seismic geologic structures. However, precise coherence attributes between adjacent seismic traces are difficult to compute due to the nonstationary and non-Gaussian property of seismic data. To describe seismic geologic structures accurately, we define a high-order spectrum-coherence (HOSC) attribute. First, we have developed a time-frequency (TF) analysis method to compute a constant-frequency seismic volume with high TF resolution, i.e., the second-order synchrosqueezing wave packet transform. Then, we developed a coherence approach by combining the mutual information (MI) calculation and coherence algorithm based on the eigenvalue computation (C3). To improve computational efficiency, we adopt the information divergence instead of the eigenvalue calculation of the C3-based algorithms. By applying our coherence algorithm to constant-frequency seismic volumes, we obtain the HOSC attribute. To test the validity of the proposed workflow, we evaluate the HOSC attribute using synthetic data. After applying our workflow to 3D real seismic data located in eastern China, the HOSC attribute characterizes seismic geologic discontinuities and subtle features clearly and accurately, such as fluvial channels and subtle faults.