Affiliations: [a] School of Computer Science and Technology, Shandong Institute of Business and Technology, Yantai, China
| [b] College of Computer Science and Technology, Jilin University, Changchun, China
Correspondence:
[*]
Corresponding author. Quansheng Dou, School of Computer Science and Technology, Shandong Institute of Business and Technology, Yantai 264005, China. Tel./Fax: +86 535 6903541; E-mail: [email protected].
Abstract: Vessel segmentation of retinal image is of great significance for medical disease diagnosis. Because of noise, non-uniform illumination etc., it is hard to precisely segment all the vessels especially small thin vessels. This paper proposed a new retinal image segmentation method called TMVD based on pixel specificity and self-adaptive classification strategy, which is implemented in three phases, in the first phase, by setting a higher pixel specificity threshold, the vessels are roughly segmented, then during the second phase, each undetermined pixels acts as an Agent, within a multi scale threshold range, the Agent revises its own status according to the status of its neighbor Agent, and then the segmentation task is completed, finally in the third phase, the noise regions in the segmentation results are cleared by double-layer window method so as to increase the segmentation accuracy. By testing TMVD on DRIVE database, the experiment shows that it is more accurate and efficient than state-of-the-art methods.
