Robust Vessel Segmentation in Fundus Images

Table 8

Accuracy comparison of different settings using the high resolution fundus database.

Algorithm	Accuracy	Absolute change
Proposed method	0.9618±0.0065	—
Without preprocessing	0.9558±0.0064	0.62%
Thresholds decreased by 1%	0.9614±0.0061	0.04%
Thresholds increased by 1%	0.9607±0.0062	0.11%
Without postprocessing	0.9401±0.0085	2.25%
Doubled morphology kernel size	0.9616±0.0060	0.02%
𝜎=2.0 for Hessian	𝟎.𝟗𝟔𝟐𝟏±𝟎.𝟎𝟎𝟔𝟐	0.03%
𝜎=3.0 for Hessian	𝟎.𝟗𝟔𝟐𝟏±𝟎.𝟎𝟎𝟔𝟏	0.03%
𝜎=4.0 for Hessian	0.9617±0.0064	0.01%