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

Table 2

The average classification accuracy rates (%), standard deviations, (%) and reconstruction errors (R_error) of the proposed approach with M=5.

Dsize	5	10	15	20	25
SVM +; random	92.01; ±2.15	94.23; ±2.47	95.57; ±2.28	95.04; ±2.63	94.80; ±2.89
SVM +; K-means	93.67; ±2.44	94.88; ±3.01	96.23; ±2.50	97.67; ±2.37	96.35; ±2.89
SVM +; ours	96.25; ±2.42	98.00; ±1.96	98.50; ±1.82	99.00; ±2.29	98.80; ±2.33
kNN +; random	91.36; ±2.17	92.75; ±2.46	94.37; ±2.86	94.50; ±2.05	94.21; ±2.49
kNN +; K-means	92.69; ±3.26	93.97; ±2.87	95.61; ±2.54	95.11; ±2.63	94.93; ±2.37
kNN +; ours	94.50; ±2.79	97.25; ±2.33	98.00; ±2.13	98.50; ±2.35	98.31; ±2.64
R_error; (random)	0.427	0.410	0.407	0.398	0.378
R_error; (K-means)	0.329	0.318	0.312	0.301	0.296
R_error; (ours)	0.118	0.116	0.108	0.105	0.094