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

Table 6

The average classification accuracy rates (%), standard deviations (%), and reconstruction errors (R_error) of the proposed approach with M=100.

Dsize	50	150	300	500	700	1000
SVM +; random	91.51; ±3.76	93.16; ±3.49	93.68; ±3.49	94.03; ±3.18	93.11; ±3.71	91.85; ±2.54
SVM +; K-means	92.56; ±3.61	94.03; ±3.55	94.36; ±3.72	94.87; ±3.54	93.96; ±3.34	92.03; ±2.72
SVM +; ours	94.15; ±3.15	94.25; ±3.19	94.75; ±3.47	95.01; ±3.06	94.00; ±3.35	92.25; ±2.94
kNN +; random	90.07; ±3.89	90.83; ±3.36	92.07; ±3.82	91.36; ±2.44	90.36; ±3.79	89.06; ±3.99
kNN +; K-means	91.11; ±3.13	91.09; ±3.21	92.31; ±3.64	92.89; ±2.30	90.87; ±3.81	89.47; ±3.74
kNN +; ours	92.50; ±3.34	91.30; ±3.51	92.50; ±3.04	93.23; ±2.98	91.00; ±3.33	89.75; ±3.36
R_error; (random)	0.486	0.479	0.468	0.456	0.439	0.435
R_error; (K-means)	0.384	0.380	0.371	0.362	0.358	0.349
R_error; (ours)	0.225	0.217	0.213	0.209	0.207	0.205