Predictive coding of depth images across multiple views