基于上下文特征的视网膜血管分割算法

针对视网膜血管分布复杂且多变,提出一种基于上下文特征提取的视网膜血管分割算法.首先通过霍特林变换(Karhunen-Loeve,K-L)生成灰度图并经过预处理增强对比度.然后经过局部信息熵进行采样.该网络编码部分的多感受野残差编码模块在兼顾速度的同时对特征进行充分的提取.同时底部的特征融合模块由非对称融合非局部模块和非...

Retinal vessel segmentation algorithm based on context characteristics

Aiming at the complexity and variability of retinal vascular distribution, a novel retinal vascular segmentation algorithm based on contextual feature extraction was proposed.Firstly, the grayscale image is generated by K-L transformation and pre-processed to enhance the contrast.Then sampling is carried out through local information entropy.The multi-receptive field residual coding module of the network coding part takes into account the speed and at the same time fully extracts the features.At the same time, the feature fusion module at the bottom is composed of asymmetric fusion non-local module and asymmetric pyramid non-local module, which is used to fuse the context features of the image.The decoding part is composed of a number of miniature U-shaped networks, which ensures the effective fusion of low-level features and high-level mapping features and the deep-level re-extraction.The algorithm in this paper was simulated on DRIVE, with accuracy of 96.45%, specificity of 98.37%, sensitivity of 82.7%, and AUC of 98.64%. The experimental results show that the algorithm can effectively segment retinal blood vessels.