Fig. 1. Basic structure of CNN

Fig. 2. Flow chart of proposed method

Fig. 3. Complex-CNN structure

Fig. 4. Accuracy and loss during training process and validation process. (a) Accuracy; (b) loss

Fig. 5. Gradient map. (a) Dark to light gradient map, used to construct the weight matrix for segmenting ILM, INL-OPL, IS-OS; (b) light to dark gradient map, used to construct the weight matrix for segmenting NFL-GCL, IPL-INL, OPL-ONL, BM

Fig. 6. Comparison of OCT retinal image segmentation results in fovea area. (a) Manually drawn result; (b) graph search method; (c) method based on CNN; (d) proposed method

Fig. 7. Comparison of OCT retinal image segmentation results with vascular shadow. (a) Manually drawn result; (b) graph search method; (c) method based on CNN; (d) proposed method