基于多特征融合和随机森林的视网膜血管分割

摘要

For the ophthalmic disease computer-aided diagnosis, this paper presents a multiple feature fusion fundus retinal blood vessels segmentation algorithm based on Random Forest. For each pixel in the field of view, a 23-D feature vector is constructed (encoding information on the moment invariant, gray level co-occurrence matrix, LoG with Gaussian second derivative, gradient of the image, phase congruency and Hessian matrix).Then a matrix is constructed for pixel of the training set as the input of the Random Forest; as a result, a Random Forest classifier used for classifying the test images is obtained. Finally, the post-processing method based on the connected area is used to make up blood vessels. The experimental re-sult testing on DRIVE database demonstrates that our method performance is better than other state-of-the- art methods based on machine learning. Meanwhile, the average accuracy, sensitivity, specificity are 0.9606, 0.7447, 0.9838, respectively.%为了进行眼底疾病辅助诊断,提出一种基于多特征融合和随机森林的视网膜血管分割方法.首先为彩色眼底图中的每个像素点提取一个23维特征向量(包括图像不变矩、灰度共生矩阵、LoG结合高斯二阶导、梯度法、相位一致性和Hessian矩阵特征);然后选取一定数量的像素点,提取其特征共同构造一个特征矩阵作为输入数据,并采用随机森林算法训练分类器;再用训练好的分类器对待分割图像中的像素点进行分类,判断其是否为血管点;最后在初步分割基础上进行基于连通区域补足血管的后处理,得到优化后的血管分割结果.在DRIVE公共数据库上进行实验的结果表明,该方法平均精确度达0.9606,平均灵敏度达0.7447,平均特异性达0.9838,比已有方法性能更优.