| INPUT: two images and n correspondences |
| //Initialization |
| (1) | vector_inx = [1:n] |
| (2) | bestMod = |
| (3) | vector\_inr = |
| (4) | minNFA = |
| (5) | nIter=(nIterRe = nIter/10) |
| (6) | nData = x1.ncol () |
| (7) | errorMax = 0 |
| | //Main estimation loop |
| (8) | for Iters = 0 to nIters do |
| (9) | vector_sp = USample (size_Sample, vector_inx) |
| (10) | vector_mods = Fit (vector_sp) |
| (11) | end for |
| | //Evaluate Models |
| (12) | better = false//Whether one of the tested models improves the NFA |
| (13) | for k = 0 to vector_mods.size do |
| (14) | error = Error (vector_mods[k]) |
| | //Residuals computation and ordering |
| (15) | for i = 0 to nData do |
| (16) | error = Error (vector_mods[k]) |
| (17) | vector_res = ErrorIndex (error) |
| (18) | sort (vector_res) |
| (19) | end for |
| (20) | end for |
| //Statistical detection of the best meaningful subset (inliers/outliers) |
| (21) | ErrorIndex.best = bestNFA |
| | //Find a better model |
| (22) | if (best.error < minNFA), then |
| (23) | minNFA = best.error |
| (24) | vector\_inr.resize (best, inx) |
| (25) | end if |
| (26) | for i = 0 to best.index do |
| (27) | vector_inr[i] = vector_res[i].index |
| (28) | errorMax = vector\_res[best.index-1].error//Error threshold |
| (29) | bestMod = vector_mod[k] |
| (30) | better = true |
| (31) | precision = denormalizeError (errorMax) |
| (32) | end for |
| | //Optimization |
| (33) | If (better and minNFA < 0) or (Iter + 1 = nIterRe), |
| then |
| (34) | vector_inx = vector_inr |
| (35) | if (nIterRe), then |
| (36) | nIter = Iter+1+nIterRe |
| (37) | nIterRe = 0 |
| (38) | end if |
| (39) | end if |
| (40) | return bestMod, vector_inr, minNFA, precision |
