I think it is not the same as the reversed process. If you generate random patterns and then test on a real curve, it will select the patterns that work with that curve. So you get curve fitted patterns that work only with that curve, not in live trading. Isnt that the opposite of what you want to do?