Image Segmentation and Classification

I have recently been tasked to a project which primarily deals with image segmentation. I am supposed to read in an MR brain image and apply k-means clustering on the image with k = 5. Obtain segmented regions through pixel classification using the clustered classes. Compare the segmented regions with those obtained from the optimal gray value thresholding method.

I know k-means clustering is not too difficult but I'm not sure how to "Obtain segmented regions through pixel classification using the clustered classes" and I would like to seek some professional advice from the community to point me in the right direction on what I should be looking at or doing.

Feel free to drop me any comments. Any help rendered is deeply appreciated.

Best Regards Westin Messer

Answer 1

[Image Analyst]

See my attached kmeans demo for a gray scale image. Adapt as needed.

By the way, kmeans is a dumb (bad) method for tumor detection. I assume it's just for an illustrative student exercise rather than a real world situation.

[Westin Messer]

Thank you very much. I am having a an issue though, when I try and run your code with your brain image I get the following error:

Error using kmeans
Too many input arguments.
Error in mathworks (line 47)
[clusterIndexes, clusterCenters] = kmeans(grayLevels, numberOfClusters,...

[Westin Messer]

I figured it out. Sorry about that.

[Image Analyst]

So did it work for you? Like I said, it is not robust so it won't detect every tumor from 0% to 100% in size. Plus any pixels with that gray level will get selected, regardless if they are actually tumor pixels or not, that just happen to have the same gray levels by chance.