algorithm - K-means 分割后如何对对象进行计数

Question

我正在从事图像分割工作，测试了很多不同的分割算法，以便进行比较研究。目前我已经实现了 k 均值算法。我想自动计算图像中分割的对象(每个图像上的粉色对象和绿色对象)。我有专家进行了手动计数，因此我想比较结果。

我的输入图像是:

我的结果图像是:

有什么方法可以让我自动化这个过程吗？

你能帮我一下吗？

提前非常感谢。约翰

Answer 1

好吧，这个方法很大程度上基于 link ImNt 发布了一些关于 matlabs 的文章 label2rgb例子。我不确定如何对区域进行聚类，但从聚类中获得的任何掩码/结果都应用作我的代码的输入

这两个示例中缺少的部分是对簇进行计数。获取聚类数量的关键是使用标签函数的输出。标签函数将所有连接的像素组标记为从 1 到 NUMBER_OF_LABELS，因此使用 max(labels(:)) 与计算所有簇的数量相同

我以完全相同的方式处理两个集群，通常我不会保留所有中间变量，它们就在这里，以便您可以在每次操作后看到输出

基本轮廓

1. Get binary mask (you can customize this)
2. fill in any holes
3. remove small regions (you can customize this)
4. label all regions (this is basically connected component grouping)
5. count regions

matlab代码

%% creates binary masks for each cluster
cluster1 = imread('cluster1.png');
cluster1 = rgb2gray(cluster1);

cluster2 = imread('cluster2.png');
cluster2 = rgb2gray(cluster2);

%im not sure how you found these clusters 1/2, but using that original mask is
%better than using this 'graythresh' Since I didn't have your original
%clustering output I tried to reconstruct it w
bw1 = im2bw(cluster1,graythresh(cluster1));
bw2 = im2bw(cluster2,graythresh(cluster2));

%alternate method for constructing binary images   
%my_gray_thresh = 0
%bw1 = cluster1 > my_gray_thresh;
%bw2 = cluster2 > my_gray_thresh;

%% the above could be replaced with the output of your clustering alg

%tries to clean up the image a bit, filling in holes
fill_bw1 = imfill(bw1,'holes');
fill_bw2 = imfill(bw2,'holes');

%takes out regions with less than smallest_area_for_dots
smallest_area_for_dots = 5; %you can play with this parameter
large_bw1 = bwareaopen(fill_bw1, smallest_area_for_dots);
large_bw2 = bwareaopen(fill_bw2, smallest_area_for_dots);

%labels the regions, this is essentially connected component analysis 
labels1 = bwlabel(large_bw1);
labels2 = bwlabel(large_bw2);

%plots the labeled image
figure(1)
subplot(2,3,[1 4]);imshow(cluster1);title('original image (grayscale)')
subplot(2,3,2);imshow(bw1);title('bw after threshold')
subplot(2,3,3);imshow(fill_bw1);title('after fill')
subplot(2,3,5);imshow(large_bw1);title('after removing small clusters')
subplot(2,3,6);imshow(label2rgb(labels1));title('labeled clusters')

figure(2)
subplot(2,3,[1 4]);imshow(cluster2);title('original image (grayscale)')
subplot(2,3,2);imshow(bw2);title('bw after threshold')
subplot(2,3,3);imshow(fill_bw2);title('after fill')
subplot(2,3,5);imshow(large_bw2);title('after removing small clusters')
subplot(2,3,6);imshow(label2rgb(labels2));title('labeled clusters')

%the number of figures is equal to the number of labels. since labels are
%numbered 1 to NUM_LABELS doing max will get us the number of dots
num_in_cluster1 = max(labels1(:))
num_in_cluster2 = max(labels2(:))

和 matlab 桌面输出

num_in_cluster1 = 243
num_in_cluster2 = 51

和结果图像