How to do pixel based classification using SVM classifier

Good day, I am graduate student and currently working on retinal blood vessel segmentation. I am new to this. I am supposed to classify which pixel is vessel pixel and which is not. I want to use SVM as a classifier for pixel based classification task. As I am a beginner, I don't know how may I train my classifier for this task and how may I get the segmented blood vessel network as an output. I am using drive database for this purpose its publicly available. 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. I will be very grateful for your valuable suggestions. Regards, Arslan

Answer 1

[Image Analyst]

The two classes you want are "blood vessel" and "fundus". Okay, fine. Now, what are the two features that you hope to identify clusters (classes) in? https://en.wikipedia.org/wiki/Support_vector_machine Gray level? Alright, seems reasonable. What else? Variation in the window? Possibly. Did you try stdfilt() or entropyfilt()? Or do you have some other feature? Did you check section 20.5 here http://iris.usc.edu/Vision-Notes/bibliography/contentsmedical.html#Medical%20Applications,%20CAT,%20MRI,%20Ultrasound,%20Heart%20Models,%20Brain%20Models for algorithms that segment out vessels from ophthalmological images?

[Arslan Ahmad]

Dear Sir, I am currently a research paper mentioned below. They used GMM classifier and I want to use SVM in classification stage. Moreover the feature vector is based on Gabor wavelet feature, entropy and Gray level. http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=1677727&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D1677727

J.V.B. Soares, J.J.G. Leandro, R.M. Cesar, H.F. Jelinek, M.J. Cree, Retinal vessel segmentation using the 2-D Gabor wavelet and supervised classification, IEEE Transactions on Medical Imaging 25 (2006) 1214–1222

[Image Analyst]

I don't have the Stats Toolbox so I can't help you with the SVM function beyond just looking at the help online, which you can do. Sorry.

[Arslan Ahmad]

Dear Sir, The stats Toolbox isn't required for this task. please suggest me some link from where I may learn pixel based classification using MATLAB and a way to train classifier for that task. I will be really grateful.

[Image Analyst]

Personally I haven't used SVM, but I would guess, based on Wikipedia, that you just take a representative sample that spans all cases you're ever likely to see and run it through to get a model. Then you use that model on all subsequent data. You might look through the stats tutorials here http://www.sportsci.org/resource/stats/

and this on sample size estimation: http://www.sportsci.org/2006/wghss.htm