Algorithm 2: Statistic filtering of feature displacement

Input: Candidate matched images $I_1$, $I_2$. Output: Good feature matches $V{M}_{good}$ 1: Detect features $I_1$, $I_2$ to obtain descriptors $M_{des1}$, $M_{des2}$ and key points $M_{keys1}$, $M_{keys2}$ 2: Match $M_{des1}$, $M_{des2}$ to obtain the original matches $V{M}_{matches}$ with brute force matcher and hamming distance 3: Calculate the key-points displacements for $V{M}_{matches}$ in $x$ and $y$ components $\Delta u=u_1-u_2$, $\Delta v=v_1-v_2$ 4: Create a 2D histogram with $\Delta u$ and $\Delta v$ to confirm the highest bins for mode approximation 5: Use the sample within the radius to perform parameter estimation of the Laplacian distribution in $x$ and $y$ 6: Determine the min and max boundary values to include a certain percentage ratio = 0.9 of inliers, assuming a Laplacian distribution 7: Find the matches $V{M}_{u\_mats}$ according to the boundary in Step 6 8: Repeat Steps 6 and 7 to find the matches $V{M}_{v\_mats}$ 9: Calculate the common element $V_{com}$ from $V{M}_{u\_matches}$ and $V{M}_{v\_matches}$ 10: For $id$ in $V{M}_{matches}$ 11:  If $id$ in $V_{com}$ 12:   Pushback the corresponding element into $V{M}_{good}$ 13:  End 14: End