Cascade Training does not work with binary images

I am working on a project and I am trying to train a cascade classifier to detect faces. The problem is that my faces are first filtered and then thresholded, so the result is a binary image with the features of the faces, such as   The training process always hangs in any stage trying to get new negative images. I have launched the process with 3000 positive images and 6000 negative images, all of them binary. I also have tried with 100 positive and 1000 negative to make a test, but it also hangs. I also have tried both Haar and LBP. Does anyone know what can I do? If I stop the process in the stage where it hangs, it doesn't detect faces propperly. Thank you very much.