Given a set of real images, Novel View Synthesis (NVS) aims to produce views of a scene that would correspond to that of a virtual camera. There exist many approaches to solving this problem. We consider physically valid NVS methods, in particular those based on epipolar and trifocal transfer. We review a number of methods and place them into a common framework of view specification and mapping method. We present a new method for directly specifying the novel camera motion for epipolar transfer. We also develop a backward mapping scheme for trifocal transfer which overcomes the problems associated with standard forward mapping methods.
Keith Connor, Ian D. Reid