Automatic Optic Disc Detection in Color Retinal Images by Local Feature Spectrum Analysis

Table 3

The average classification accuracy rates (%), standard deviations (%), and reconstruction errors (R_error) of the proposed approach with M=10.

Dsize	10	30	50	70	90	100
SVM +; random	94.07; ±2.96	94.79; ±2.78	95.14; ±2.70	96.88; ±2.87	96.62; ±2.34	96.51; ±3.06
SVM +; K-means	95.63; ±2.45	95.77; ±2.62	96.21; ±2.33	97.96; ±2.91	98.71; ±1.99	97.60; ±2.98
SVM +; ours	98.50; ±2.10	98.75; ±1.32	99.20; ±1.93	99.50; ±2.06	99.75; ±1.79	99.65; ±1.96
kNN +; random	93.89; ±3.12	94.20; ±2.09	95.53; ±2.96	96.34; ±1.98	96.55; ±2.37	96.43; ±2.55
kNN +; K-means	94.22; ±2.93	95.23; ±2.68	96.60; ±2.45	97.59; ±2.85	98.21; ±2.14	97.83; ±2.22
kNN +; ours	97.25; ±2.13	97.25; ±2.27	98.50; ±2.35	98.64; ±2.13	98.80; ±2.25	98.58; ±2.38
R_error; (random)	0.449	0.437	0.423	0.415	0.408	0.396
R_error; (K-means)	0.334	0.326	0.320	0.316	0.306	0.297
R_error; (ours)	0.181	0.176	0.162	0.153	0.140	0.138