K-means clustering

2 次查看（过去 30 天）
Radhika Bhagwat
Radhika Bhagwat 2017-1-21
评论： Image Analyst 2018-2-8
I am doing color based segmentation using kmeans clustering.I am using inbuilt function of matlab(kmeans).My input image has object and background where I need to segment the object. For that I am using cluster value as 2 and repeating the clustering 3 times.The problem I am facing is that for some images, the output of k-means is very bad the first time, but when I try doing the segmentation for the 2nd time it gives me good results. Why is this happening?Is it because of the light variations in the image? Posting the original image, image with wrong segmentation and image with right segmentation

请先登录，再回答此问题。

回答（3 个）

Image Analyst
Image Analyst 2017-1-22
See my attached demo for doing kmeans clustering on RGB images:
  7 个评论
显示 5更早的评论
Srinivas Reddy
Srinivas Reddy 2018-2-8
It's working...Thanks a lot
Image Analyst
Image Analyst 2018-2-8
For hands1 you should have indexes = 76800 x 1, rows = 240, and columns = 320. You must have changed something, but I'm glad you restored it and got it working again.

请先登录，再进行评论。

Image Analyst
Image Analyst 2017-1-21
Well obviously there are not 2 clusters. There are 3 dominant colors: green, brown, and black. Use k=3 in your code and it should improve.
Better yet, if you know you are going after certain colors like green, do thresholding in HSV color space. Try the color thresholder on the Apps tab of the tool ribbon.
  4 个评论
显示 2更早的评论
Radhika Bhagwat
Radhika Bhagwat 2017-1-22
Here is the code:
I= rgb image;The first image posted earlier.
cform = makecform('srgb2lab');
lab_he = applycform(I,cform);
ab = double(lab_he(:,:,2:3));
nrows = size(ab,1);
ncols = size(ab,2);
ab = reshape(ab,nrows*ncols,2);
nColors = 2;
[cluster_idx, cluster_center] = kmeans(ab,nColors,'distance','sqEuclidean', ... 'Replicates',3);
pixel_labels = reshape(cluster_idx,nrows,ncols);
figure();imshow(pixel_labels,[]), title('image labeled by cluster index');
After executing this code, the output I get for the first time is the second image that I have posted earlier. But if I try executing this entire code again I get a perfect segmented image (the 3rd image posted earlier). My question is why is this happening?(ie some times it gives bad segmentation while some times a good one). I understand that k-means selects the seeds randomly(So I could get 1 for object and 0 for background or vice-versa and this could change on every run).But I think ,the segmented part should remain same for every run(ie object and background). No matter if object/background is 1 or 0.But in the second image the algorithm has given the leaf and the shadow as object.While on second run,I get perfect result.
Image Analyst
Image Analyst 2017-1-22
Did you run my kmeans demo I made up for you? It's in my second answer on this page. It plots the 3-D color gamut. Here is what a scatterplot of your a,b data looks like looking down the L axis:
and here is what it looks like from the side:
Do you see 2 well defined, well separated clusters there? No, you do not. The colors go continuously from one color to the next. There are going to be some colors that are "in between" colors and some maybe classified as one thing and some as the other thing, perhaps in disagreement with what you thought they should be.
That is why after doing color classification, by whatever method, often/usually you need additional steps to clean things up.
For what it's worth, I'm attaching another statistical method demo given to me by the Mathworks. It uses principal components analysis.
Additionally there are thresolding-based methods of color segmentation in my File Exchange: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

请先登录，再进行评论。

steny ynets
steny ynets 2017-8-28
what is the code to differentiate diseased and healthy part of the leaf

类别

在 Help Center 和 File Exchange 中查找有关 k-Means and k-Medoids Clustering 的更多信息

标签

产品

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by