206
M. Wu et al.
methods, our algorithm has several advantages: (1) A two-stage strategy combines
the global and local information for optic disc segmentation. In coarse disc margin
location, we consider the structural characteristics of ONH for initial NCO detection.
In the patch searching procedure, we seek to find the most likely patch centered at
the NCO near NCO candidates. (2) Unlike A-scan classification-based segmentation
methods that directly classify every column in B-scan images as disc (rim or background), our algorithm only searches in a restricted region, increasing efficiency.
Furthermore, the SVM classifier spends less computational time than a k-NN classifier because it can use pre-training model data, further boosting efficiency. (3) As
shown in Fig. 8.5, the segmentation results of the proposed algorithm have smoother
Fig. 8.6 Comparisons of C/D ratio quantification by different algorithms. a Area C/D ratio quantification. b Error of C/D ratio evaluation
M. Wu et al.
methods, our algorithm has several advantages: (1) A two-stage strategy combines
the global and local information for optic disc segmentation. In coarse disc margin
location, we consider the structural characteristics of ONH for initial NCO detection.
In the patch searching procedure, we seek to find the most likely patch centered at
the NCO near NCO candidates. (2) Unlike A-scan classification-based segmentation
methods that directly classify every column in B-scan images as disc (rim or background), our algorithm only searches in a restricted region, increasing efficiency.
Furthermore, the SVM classifier spends less computational time than a k-NN classifier because it can use pre-training model data, further boosting efficiency. (3) As
shown in Fig. 8.5, the segmentation results of the proposed algorithm have smoother
Fig. 8.6 Comparisons of C/D ratio quantification by different algorithms. a Area C/D ratio quantification. b Error of C/D ratio evaluation