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

Table 5

The average classification accuracy rates (%), standard deviations (%), and reconstruction errors (R_error) of the proposed approach with M=50.

Dsize	50	150	300	500	700	1000
SVM +; random	91.56; ±2.98	94.76; ±3.77	93.49; ±3.16	94.27; ±2.75	93.17; ±3.73	93.01; ±2.39
SVM +; K-means	92.98; ±2.75	95.32; ±3.39	94.76; ±3.64	95.19; ±2.63	94.24; ±3.85	93.94; ±2.73
SVM +; ours	94.75; ±2.53	95.48; ±3.46	96.50; ±3.10	95.45; ±2.12	94.42; ±3.44	94.40; ±2.53
kNN +; random	90.13; ±3.17	91.83; ±3.94	94.15; ±3.19	91.37; ±2.34	91.01; ±3.73	90.07; ±2.39
kNN +; K-means	90.87; ±3.61	92.09; ±3.46	95.09; ±3.90	91.88; ±2.65	90.35; ±3.94	90.78; ±2.31
kNN +; ours	91.95; ±3.11	92.25; ±3.77	96.25; ±3.56	92.00; ±2.89	91.75; ±3.13	91.25; ±2.71
R_error; (random)	0.472	0.465	0.459	0.447	0.436	0.429
R_error; (K-means)	0.375	0.367	0.358	0.344	0.329	0.320
R_error; (ours)	0.194	0.187	0.184	0.179	0.177	0.174