Feature extraction by using gabor filter

Hello,

I'm quite new in the field of SVM's and would need an advice. I currently trying to optimize a SVM (based on LIBSVM). I was thinking about using a gabor filter to extract the features from my images. The problem is that this topic is quite new for me and would need an advice how I should deal with the data generated after applying my filter in order to use them to train my SVM. After I had applied my filter I have many responses (real and imaginary part )depending on how many filter and how many orientations I using.

The question is how should apply this data to my SVM. Should I use only one response? if so, which one is the best? Should merge the responses? if so, how? How should deal this the real and imaginary part of the response? Should I apply an fft to this data?

Thank you in advance & cheers Gregor

P.s the Gabor filter is based on based on Gabor filter toolbox; Jarmo Ilonen and Joni Kämäräinen