hello Sheema,

Are you sure that SVM is really the central point of the problem ? Basically, SVM is a method for finding groups in a supervised way That means that you must have at least a learning set formed of vectors . For each vector of this set, you must have an a priori knowledge giving the belonging to its qualitative group. In your case, I cannot see what is this learning set. It can be:

1) A set of vector pixels for which you know the associated label (for example : this pixels belongs to a shape of kind numbered XXX), in which XXX is the group.

This can be appropriate if you have multivariate or hyperspectral images. If your images are RGB ones, it is perhaps irrelevant to use SVM with learning vectors including only 3 elements (the R, G, B intensities). If possible, try to use an approach involving thresholding, if your shape has a different color than the background.

2) A set of shapes that you have isolated. In this case, for each of them, you have computed features (for example : length, width, surface, histograms of color characteristics , shape features). The learning set is here composed of these vectors of features For each vector , you must be able to atrtribute a qualitative group.

Before trying SVM, I would recommand to try simple linear methods such as PLS discirminant analysis or (even) basic discriminant analysis.

You could find many information and publications on my reaseargate pages.

("Domiinque Bertrand")

Kind regards,

Dominique

Thank you all.In my dataset each image has a person with infected ,healthy skin and background ?