Fault picking is a critical, but human-intensive component of seismic interpretation. In a bid to improve fault imaging in seismic data, I have applied a directional Laplacian of a Gaussian (dLoG) operator to sharpen fault features within a coherence volume. I compute an M by M matrix of the second moment distance-weighted coherence tensor values that fall within a 3D spherical analysis window about each voxel. The eigenvectors of this matrix define the orientation of planar discontinuities while the corresponding eigenvalues determine whether these discontinuities are significant. The eigenvectors, which quantify the fault dip-magnitude and dip-azimuth, define a natural coordinate system for both smoothing and sharpening the planar discontinuity. By comparing the vector dip of the discontinuity to the vector dip of the reflectors, I can apply a filter to either suppress or enhance discontinuities associated with unconformities or low signal-to-noise ratio shale-on-shale reflectors. Such suppression become useful in the implementation of subsequent skeletonization algorithms. Automatic fault picking processes for accelerated interpretation of basins also become much easier to implement and more accurate. I demonstrate the value and robustness of the technique through application to two 3D post stack data volumes from offshore New Zealand, which exhibit polygonal faulting, shale dewatering, and mass-transport complexes. Finally, I use these filtered faults as input to an ant-tracking algorithm and automatic fault extraction and find significant improvement in the speed and accuracy of fault interpretation.