Feature-Based Retinal Image Registration Using D-Saddle Feature

Table 5

Details of success rate and registration accuracy of successful registration for GDB-ICP, Harris-PIIFD, H-M, Saddle, and D-Saddle in FIRE dataset.

Methods	Nsuccess	Success rate (%)	Registration accuracy (TRE) of successful registration (pixels)					Std. deviation of all pairs
			Mean	95% confidence interval of the mean		Min	Max
				Lower	Upper
Overall
GDB-ICP	37	28%	1.988	1.566	2.411	0.486	4.952	1.479
Harris-PIIFD	5	4%	2.573	0.571	4.576	0.785	4.244	365.754
H-M	22	16%	1.232	0.849	1.616	0.554	3.315	39.918
Saddle	21	16%	2.059	1.299	2.818	0.719	4.977	7.553
D-Saddle	57	43%	2.329	1.862	2.797	0.596	4.993	3.988
Category (super-resolution)
GDB-ICP	17	24%	0.783	0.703	0.863	0.486	0.988	0.778
Harris-PIIFD	2	3%	0.846	0.071	1.621	0.785	0.907	1.969
H-M	18	25%	0.839	0.780	0.897	0.554	0.995	1.886
Saddle	13	18%	0.858	0.806	0.911	0.719	0.963	1.265
D-Saddle	32	45%	0.852	0.810	0.894	0.596	0.999	0.897
Category (image mosaicking)
GDB-ICP	16	33%	3.068	2.620	3.516	1.946	4.952	1.134
Harris-PIIFD	0	0%	N/A	N/A	N/A	N/A	N/A	580.486
H-M	0	0%	N/A	N/A	N/A	N/A	N/A	18.980
Saddle	3	6%	4.921	4.785	5.058	4.867	4.977	4.532
D-Saddle	13	27%	4.520	4.271	4.769	3.741	4.993	3.301
Category (longitudinal study)
GDB-ICP	4	29%	2.792	1.903	3.680	2.354	3.603	3.444
Harris-PIIFD	3	21%	3.725	2.551	4.899	3.319	4.244	396.753
H-M	4	29%	3.004	2.664	3.343	2.857	3.315	110.853
Saddle	5	36%	3.462	2.495	4.429	2.767	4.769	4.277
D-Saddle	12	86%	3.896	3.303	4.490	2.534	4.906	3.637

Standard deviation of TRE from all image pair (both failed and successful registration).