High feature selection with AdaboostM1 in Matlab to reduce computational complexity

Question

I'm implementing an algorithm in Matlab which test accuracy for detection of modified images. The accuracy is provided by an SVM. But my problem is how to select high features with adaboost with the purpose to reduce computational complexity.

In detail, I have a feature vector (1xN) for an image and I'm not sure that I can use a similar vector in adaboost. I'm using fitensemble function in Matlab with this vector as X, numberens= 30,50,75,100, 'tree' as parameters. I should get D-dimensional new feature vectors. My question is if observations in my data input are the features and so, I should convert the row vector in a column vector, or am I completely wrong?

Where can I find the new feature vector or how can I obtain it in the resulting model, ens?

Details: "Boosting Feature Selection can be used to select optimal features to reduce the computational complexity. After D iterations, we can get D-dimensional new feature vectors, then they will be used for final classification." What I think is that my feature values may be observations and I have only one column, related for that image. Adaboost should return me a new feature vector with dimension D. For example: if the length of my initial feature vector, F, is 256, should I get a new feature vector of size 30, in the case of using 30 iterations? F contains many zeros and some values where maximum can be something equal to 100000.