Should I try the classifier on flipped images too?

edit

I am training a Bag Of Words classifier. My application is based on this code. I have a question linked to the training database:

Should I mirror flipped the images, or this kind of classifier is invariant to the flipping operation?

For example, I have a class of car_lateral. Shall I mirror flip the images and add both (initial and flipped) in the training database?