If I want to implement a cascade classifier for detecting e.g. my jacket on an image, I can obtain the positive samples for it like this: take pictures of it from different angles and put each picture on a few backgrounds. If I want to detect any jacket (not only the mine one), I would need to take photos of ~500 (or more for a more precise classifier) different jackets.
But I'm not sure if there's a similar approach for negative samples. So, my question: is there a general approach (like the above one) to obtaining negative samples? Or do I have to craft one for each classifier?
I'm not seeking recommendations for any specific books, tutorials, tools etc. I only want to know if there's a generalized approach for obtaining negative samples.