如何在视网膜图像中找到视神经 [英] How can I find optic nerve in a retina image

问题描述

我正在使用matlab R2011b所以我没有'imfindcircles'的功能。我的目标是检测视神经并使其完全变白。我该怎么办？原始图片：https：//www.dropbox.com/s/btzxkt778u5cga6/A.jpg？dl = 0 [^]二进制图片：https：//www.dropbox.com/s/t5jpw505lx44u81/ B.jpg？dl = 0 [^]在结果图像中，红色圆圈中的区域应为白色：https：//www.dropbox.com/s/wgn5vhmpczk77p4/C.jpg？dl = 0 [^]

解决方案

我没有使用Matlab，但我已经使用AForge.NET认识到了一些功能，可以告诉你，你的情况并不那么容易，因为你形象的几个问题：缺乏色彩;你正在寻找的地方是模糊的，它的形状是非常不规则的，因为倾斜的灯光显示出一些静脉的阴影;图像中还有其他模糊点;它们看起来与你的眼睛不同，但会混淆标准的识别算法;有长白色的静脉也可以被检测为假阳性。



您需要进行某些实验来解决这个图像。但一般的方法应该是这样的：首先，你对图片进行对比变换，可能会走向极端。然后执行blob分析。如果斑点检测算法具有尺寸区分（按尺寸过滤），尤其是在两个方向（垂直，水平）上，则应使用它。在blob检测之后，我几乎肯定你会有一些误报，所以你必须执行自己的过滤以选择正确的过滤。只有一个对象这一事实大大简化了这个问题。最有可能的是，这个特征的大小位于一个相当狭窄的区域，您是否可以查看检测到的斑点的凸包实例，并找到一个最圆的，其尺寸最接近预期的大小。



如果我的建议不够明确，很抱歉。我需要自己进行大量的研究以解决这种认识，但我觉得这很有可能。

-SA

Hi, I am using matlab R2011b so I have't 'imfindcircles' function. My aim is to detect optic nerve and make it totaly white. How can I do that?Original image:https://www.dropbox.com/s/btzxkt778u5cga6/A.jpg?dl=0[^]Binary image:https://www.dropbox.com/s/t5jpw505lx44u81/B.jpg?dl=0[^]In the result image the area in the red circle should be white:https://www.dropbox.com/s/wgn5vhmpczk77p4/C.jpg?dl=0[^]

解决方案

I'm not using Matlab, but I've done recognition of some features using AForge.NET and can tell you that your case is not the so easy, due to several problems of your image: the lack of colors; the spot you are looking for is fuzzy, it's shape is very irregular due to oblique lighting showing shadows from some veins; there are other fuzzy spots in the image; they look different to your eyes but will confuse the standard recognition algorithms; there are long whitish veins which also can be detected as false positives.

You will need certain experimenting to "solve" this image. But the general way should be this: first, you perform contrasting transform to the picture, probably going to extreme. Then you perform blob analysis. If the blob detection algorithm has size discrimination (filtering out by size), especially in two directions (vertical, horizontal), you should use it. After blob detection, I'm almost sure you are going to have some false positives, so you will have to perform your own filtering to choose the right one. This problem is greatly simplified by the fact that you have only one object. Most likely, the sizes of this feature lie in a pretty narrow domain, do you can look at the convex hull instances of the detected spots and find one which is the "most rounded" with the sizes closest to the expected size.

Sorry if my advice is not definitive enough. I would need to conduct a good deal of research myself to resolve this kind of recognition, but I feel that it's quite possible.

—SA

这篇关于如何在视网膜图像中找到视神经的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！