Prediction and characterization of irregular crack geometrical characters from detection signals are major challenges for electromagnetic non-destructive evaluation (ENDE). In this paper, a new approach using spatial leakage magnetic field vectors in three mutually perpendicular coordinate planes is proposed for characterization of crack geometrical features considering defect surface profile, orientation and propagation angles. New signal features of B_α, B_β and B_γ extracted from leakage magnetic field in three coordinate planes of X-Y, X-Z and Y-Z are provided to improve detection sensitivity and capability for crack geometrical features evaluation. Sample including four irregular surface-breaking cracks is employed and finite element method (FEM) simulations of magnetic flux leakage (MFL) in 3D are carried out to analyze leakage magnetic field distribution and verify feasibility of the proposed approach. This approach shows competitive ability for crack shape, orientation and propagation features recognition compared with three axis components in 3D MFL. The validation of this approach for natural RCF cracks evaluation and crack dimension reconstruction is scheduled to be done in a near future.