Algorithm 2 The pseudo-code of refinement

Input: initial clusters partition   the normal of each point $\mathit{N}$,   the normals of orientation planes $Plane$,   the number of orientation planes numPlanes,   the number of points numPoints   the maximum number of iterations maxNumIters   the impact factor λ     the number of neighbor voxels used for the search nnFilledVoxels   voxel list V Output: updated clusters $partition$   $\omega =\lambda /numNeighbors$   For $i=0$ to numPoints-1 do     For $j=0$ to numPlanes-1 do       scoreNormal[i][j] = $N[i]·Plane[j]$     End For   End For   For $k=0$ to maxNumIters -1 do     For $i=0$ to $V.size$-1 do       voxScoreSmooth[$V_i$][ ] = 0       For $j=0$ to $V_i.size$-1 do         voxScoreSmooth[$V_i$][$partition[V_i[j]]$]++//calculate the voxel smooth score       End For     End For     For $i=0$ to numPoints-1 do       For $j=0$ to numPlanes-1 do         $v=filledVoxels[i]$//the voxel at which the i-point is located         Find the neighboring voxels and store them in nnFilledVoxels         scoreSmooth[i][j] = 0         For $n=0$ to nnFilledVoxels.size-1 do           scoreSmooth[i][j] += voxScoreSmooth[nnFilledVoxels[n]][j]         End For       score[j]= scoreNormal[i][j] + $\omega$•scoreSmooth[i][j]       End For     Cluster the point to j which related score is the highest     $tmpPartition[i]=clusterIndex$     End For     $partition[i]=clusterIndex$   End For