活动轮廓作为一种重要的图像分割工具, 近几年来在理论和应用方面都有很大的发展。然而, 现有轮廓模型在处理灰度均匀性较差的图像时, 通常存在较高的分割误差, 并且对初始轮廓曲线位置敏感。为此, 本文提出一种基于血管特征约束的活动轮廓模型, 该模型首先使用局部相位(Local Phase)的血管增强算法对图像进行增强处理以生成一种不同于图像灰度的血管特征信息, 然后将血管信息和图像灰度以线性加权的形式引入到局部二值拟合(Local Binary Fitting, LBF)能量泛函中, 指导图像血管分割。基于视网膜血管图像数据(Digital Retinal Images for Vessel Extraction, DRIV)的实验显示: 该模型能成功地从灰度分布不均匀和弱边界轮廓的视网膜图像中提取血管, 分割灵敏度和准确性分别达到74.43%和9367%, 同时对初始轮廓曲线位置的敏感性大为降低。由上述可知, 该模型具有高分割准确性和低初始位置敏感性。

Active Contour Models are the essential instruments for image segmentation and have great development in both theory and application. However, these existing models cannot work well in the presence of intensity heterogeneity, and are in general sensitive to initial curve places. Therefore, a novel active contour is proposed so as to extract correctly vessels, which takes into account image intensity and vessel features simultaneously. The new model is obtained by local phase vesselness-enhanced algorithm, and is based on local binary fitting. Experimental results, which based on publicly available Digital Retinal Images for Vessel Extraction (DRIVE), show that our model can successfully extract the desired vessels in the presence of intensity inhomogeneity with 74.43% pixel sensitivity and 93.67% segmentation accuracy, and the model is insensitive to initial curve placement. This demonstrates our model is competent for image segmentation with high accuracy and robustness.