Retinal vessels segmentation method based on dynamic threshold neural P systems with orientation feedback

The correct segmentation of thin blood vessels has always been the difficulty of retinal fundus image segmentation, because there are some problems, such as poor contrast between thin blood vessels and background, broken structure, noise, etc., which makes the convolution-based feature extraction depth network and traditional feature extraction algorithms have poor segmentation effect on thin blood vessels. Therefore, we propose a dynamic threshold neural P system model with orientation feedback (OF-DTNP systems), which relies on the local orientation of retinal blood vessels calculated by the orientational vector fusion (OVF) method. In the iterative process of the OF-DTNP system, the orientation estimation and feedback operations are performed alternately, so that the thick and thin blood vessels are enhanced. This model is tested on the DRIVE and STARE public datasets and compared with nearly 25 methods in the field of retinal fundus image segmentation. The experimental results show that this model has clear advantages in the segmentation of thin blood vessels.