I am desperatley trying to avoid rotoscoping.
No way. In a world full of choices you made the worst thinkable decision of shooting it the way it is - low contrast, many similarly colored items, fine structures. And did I mention that the refractions from that glass lamp alone will add a level of complication? There simply is nothing here that would allow even the smartest automatic effect to detect anything and masking everything manually is pretty much the only way, which if little ginger head isn't moving around too much might not be that complicated, after all.
Mylenium