Why are support vectors that receive from the Classification Learner application different from training set?

Hello, I used Classification Learner application on MATLAB to training my SVM model which I want to classify the data into 4 classes by using one-vs-all strategy. Finally, I will receive 4 decision function for each class. However, I have no clear that why the model that I extract from the app provides different values from the training set that actually the support vectors should be some sample of training data (a subset of training data) that lie on boundary parallel hyperplanes. Below shows support vectors from one class and my training data.