Here CNN based segemntation of Fundus Images is done to get the hard exudates for detection of diabetic retinopathy. The dataset used is IDRiD containing 81 fundus images.
32 x 32 patches were made for for each image and the state of central pixel was assigned as the target output.
8 layer CNN was made to classify central pixel of image patches and the finally the patch classifiers were put together.
Ground truth :
Predicted with 32x32 patch size :
