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

Table 4

The average classification accuracy rates (%), standard deviations (%), and reconstruction errors (R_error) of the proposed approach with M=30.

Dsize	50	150	300	500	700	1000
SVM +; random	92.95; ±3.64	95.79; ±3.86	95.09; ±2.14	94.36; ±2.75	94.17; ±2.49	94.02; ±2.16
SVM +; K-means	93.83; ±3.33	97.02; ±3.64	97.36; ±2.88	96.24; ±2.69	96.03; ±2.72	95.83; ±2.99
SVM +; ours	97.75; ±3.10	98.72; ±3.29	97.75; ±2.67	97.00; ±2.71	96.50; ±2.84	96.50; ±2.76
kNN +; random	92.14; ±3.69	96.86; ±3.91	93.04; ±2.80	92.96; ±3.87	92.36; ±3.60	90.05; ±3.49
kNN +; K-means	92.96; ±3.59	97.50; ±3.55	93.76; ±2.73	93.45; ±3.93	92.61; ±3.47	91.34; ±3.91
kNN + ours	96.50; ±3.17	98.00; ±3.06	94.00; ±2.89	93.75; ±3.04	92.75; ±3.21	91.75; ±3.09
R_error; (random)	0.463	0.448	0.440	0.436	0.429	0.418
R_error; (K-means)	0.356	0.349	0.336	0.328	0.319	0.311
R_error; (ours)	0.183	0.178	0.174	0.169	0.168	0.162