Extracting Blood Vessels in Retinal Images by Adaptive Thresholding
一种视网膜血管自适应提取方法

In terms of the special gray distribution in retinal images, a novel blood vessel extraction method based on adaptive thresholding is proposed in this paper. The whole image is divided into many small sub-images with identical dimension, and u threshold is calculated respectively in each sub-image for segmenting local blood vessels. Because both vessels and background are locally uniform in retinal images, there must be a threshold which is able to segment vessels precisely in a certain sub-image. The method employed for determining the local threshold not only allows sub-images to be very small, but also ensures the threshold to be optimal in the sense of least square error. A new edge tracking algorithm based on zero-crossing edge detection technique is applied in the process of threshold computing. Further more, a feature synthesis method based on region growing is presented, which is used to clear fragments in results of adaptive thresholding. The experiments on many retinal images indicate that this blood vessel extraction method is computational efficient and can extract most blood vessels including very small blood vessels.