蛇：活动轮廓模型 [英] SNAKES: Active Contour Model

问题描述

我从这里获得了Snakes算法的代码（在MatLab中实现）

http://www.mathworks.com/matlabcentral/fileexchange/28109-snakes-active-contour-models

当给它轮廓周围的初始索引时，它运行完美。但是，不幸的是，这不是我想要的。

想象一下，有一座山，我想检测出它的轮廓。但是，我只有山顶的索引。因此，初始索引是围绕该像素的索引。但是当运行算法时，蛇会越来越小直到消失。

我希望蛇长大直到找到轮廓。那可行吗？

解决方案

我不是专家，但是我已经对该主题做了一些阅读。据我了解，许多蛇算法在没有图像强迫的情况下往往会收缩，因为它们会惩罚一阶导数（| x’| ^ 2的整数），而会无意惩罚区域。

如果您可以访问它，他们将在本文中讨论这个问题，并尝试通过在成本函数中添加一个体积项来对其进行更改，以使其得到扩展。 / p>

http：/ /www.springerlink.com/index/10.1007/s00791-012-0178-8

希望有帮助。

I got the code of Snakes algorithm from here (Implemented in MatLab)

http://www.mathworks.com/matlabcentral/fileexchange/28109-snakes-active-contour-models

when you give it the initial indices surrounding the contour, It runs perfectly. but, unfortunately that isn't what I want.

Imagine that there is a mountain, I want to detect it's contour. But, I only have the index of the top of the mountain. So, the initial indices are the indices surrounding this pixel. But when running the algorithm, the snake is getting smaller and smaller till vanishing.

I want the snake to grow up till it founds the contour. Is that feasible?

解决方案

I'm not an expert, but I have done a little reading on this topic. From what I understand many snake algorithms tend to shrink in the absence of any image forcing because they punish the first derivative (the integral of |x'|^2) and that inadvertently punishes area.

If you can access it, they talk about this problem in this paper and try and alter it to get an expanding snake by adding a volume term to the cost function.

http://www.springerlink.com/index/10.1007/s00791-012-0178-8

Hope that helps.

这篇关于蛇：活动轮廓模型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！