A supervised method for retinal image vessel segmentation by embedded learning and classification

摘要

This paper presents a supervised method for blood vessel segmentation in digital retinal images by a combination of learning and classification. For an image, the method defines and computes pixel strength as primary features for conservatively computing vessel and background pixels as preliminary segmentation, from which the main segmentation selects training data to learn a neutral network (NN) classifier on the fly. Each pixel in the training data set is represented by an 8-D vector composed of intensity descriptor and pixel strength features, and the learned classifier for the image is next applied to classify the undetermined pixels. The segmentation results are further refined by filtering out the outliers. The method was evaluated on the publicly available DRIVE database, and the results showed better or comparable performance when comparing with other existing solutions in literature. The much better sensitivity and robustness of our approach with different image conditions make it potentially suitable for clinical applications such as automated screening for early diabetic retinopathy detection, and auto-and semi-automatic grading of diabetic retinopathy.