基于BP网络的人眼组织切片图像角膜分割

Cornea Segmentation in Human Eye Tissue Slice Images Based on BP Neural Network

角膜是人眼的一个关键部分,了解角膜的精细结构具有十分重要的意义.本文提出了一种基于BP(Back-propa-gation)网络的,针对组织切片图像序列提取角膜局部信息的新方法.首先对切片序列进行预处理,截取包含角膜信息的原1/2图像,其次选取预处理过的图像序列中角膜颜色差异的若干图片做训练样本,以人工分割后的图像作为导师信号,通过分析选取6个图像特征作为输入向量,对三层BP神经网络进行训练.训练完成后,采用该网络对图像序列进行分割,最后对分割图像进行后处理包括区域分割、孔洞填充及恢复图像大小.实验结果表明,该方法能够获得较好的图像序列分割效果,为角膜精细形态模型的建立及其后续研究打下基础,同时也为眼内其他重要结构的分割提供一种参考方法.

Cornea is one of the most important parts of human body＇s eyes. According to the particular obtainment of the human eyes tissue images,in this paper, a method to extract cornea＇s local information based on back-propagation （BP） neural network is proposed. First, we capture half of the images including cornea information. Then several images with different cornea colors are chosen as training samples. Manual segmentation images of training samples are used as teacher signals. The network＇s input vectors are 6 image features. We can train the BP neutral network with 3 layers and capture the segmented images. Finally, we get the exactly cornea information by region growing with the segmented images and filling holes of images. This method is turned out to get expected purpose and operate little subjective. It is a reference method to segment the other parts of eyes of these tissue slice images,and also is a basis for the research of modal model of cornea.