Learning-Based Visual Saliency Model for Detecting Diabetic Macular Edema in Retinal Image

Figure 2

Illustration of learning process and saliency computing process. Feature Property: a set of low-level visual features are extracted from some training images. Feature vectors corresponding to the top 20% (bottom 50%) of the ophthalmologists’ precise diagnoses (ground truth) are assigned +1(0) labels. Then a SVM classifier is trained from these features and is used for predicting DME on a test image. Position Property: we used a statistical analysis method to obtain it from ground truth. Finally, we combine Feature Property and Position Property adopting.