A Method for Estimating View Transformations from Image Correspondences Based on the Harmony Search Algorithm

Table 3

Inlier detection comparison in terms of the detection rate (DR), the error (), and the number of function evaluations (NFE) for standard RANSAC [14], the MLESAC [20], the SIMFIT method [21], the projection-pursuit algorithm [22], the TSSE [23], the PSO algorithm (PSO-RANSAC) [54], and the proposed HS-RANSAC approach, considering the four test images shown in Figures 5, 6, 7, and 8.

Image	Method	Detected inliers (NofI)	Missing	False alarms	DR (%)		NFE
(A) Total number of inliers (86)	Standard RANSAC	41	45	21	47.7	4.75	876
	MLESAC	55	31	14	63.9	3.11	852
	SIMFIT	62	24	11	72.0	2.98	842
	Projection-pursuit	58	28	12	67.4	3.53	798
	TSSE	48	38	14	55.8	3.42	815
	PSO-RANSAC	75	11	8	87.2	1.68	491
	HS-RANSAC	82	4	5	95.3	0.88	396
(B) Total number of inliers (72)	Standard RANSAC	32	40	18	44.4	3.98	765
	MLESAC	40	32	14	55.5	3.43	825
	SIMFIT	58	14	8	80.5	2.87	891
	Projection-pursuit	47	25	12	65.2	3.12	759
	TSSE	43	29	16	59.7	3.47	786
	PSO-RANSAC	63	9	5	87.5	1.51	374
	HS-RANSAC	70	2	3	97.2	0.79	328
(C) Total number of inliers (56)	Standard RANSAC	24	32	15	42.8	2.96	689
	MLESAC	27	29	11	48.2	2.41	628
	SIMFIT	42	14	9	75.0	1.98	724
	Projection-pursuit	37	19	13	66.0	2.85	754
	TSSE	32	24	14	57.1	2.74	776
	PSO-RANSAC	48	8	9	85.7	0.94	349
	HS-RANSAC	53	3	5	94.6	0.25	272
(D) Total number of inliers (122)	Standard RANSAC	62	60	22	50.8	4.02	832
	MLESAC	77	45	18	63.1	3.41	924
	SIMFIT	90	32	13	73.7	2.86	845
	Projection-pursuit	75	47	19	61.4	3.52	914
	TSSE	76	46	21	62.2	3.73	887
	PSO-RANSAC	110	12	10	90.1	1.41	427
	HS-RANSAC	115	7	5	94.2	0.51	338