Automatically identifies the product outline in the input image, separates it from the background and returns a foreground product image (4 channels) with the background removed, suitable for single/multiple product, complex background scenarios, etc.