Rendering a virtual view from a single reference view and its associated depth map stream suffers
from the disocclusion or exposure problem, where some regions in the virtual view have no mapping
because they were invisible in the reference view. These regions are known as holes and require applying
a filling algorithm that interpolates the value of the unmapped pixels from surrounding areas.
This disocclusion effect increases as the angular distance between the reference view and the virtual
view increases. Virtual views may be synthesized more correctly if two or more reference views, from
both sides of the virtual view, are used [Gotfryd et al. 2008]. This is possible because areas which are
occluded in one of the reference views may not be occluded in the other one. An illustrative example
of this process is given in Figure 8. Each of the reference views on the left and the right have two
components: a texture image and a depth map. Using the depth map and the corresponding camera
parameters of the reference view, as well as the camera parameters of the target view, the texture
components of the references are individually warped using a DIBR algorithm to the target view. By
blending the two resulting images, an synthetic image is generated where texture information for
occluded regions in either reference are obtained from the other