Multi view synthesis method and display devices with spatial and inter-view consistency
An embodiment of the invention is a method that provides common data and enforces multiple view consistency. In a preferred embodiment, a coordinate alignment matrix is generated after initial warped views are formed. The matrix is used efficiently to repair holes/occlusions when synthesizing final views. In another preferred embodiment, a common background layer is generated based upon initial warped views. Holes are repaired on the common background layer, which can then be used for synthesizing final views. The cost of the latter method is sub-linear, meaning that additional views can be generated with lesser cost. Preferred methods of the invention also use a placement matrix to generate initial warped views, which simplifies warping to a lookup operation.
1. A method for rendering multiple view data from a stereoscopic video data stream, comprising:
receiving or estimating a disparity map D from the stereoscopic data;
forming a placement matrix P to warp known pixels onto an image plane of a desired view, the placement matrix at position α in the placement matrix is defined as:
P k α ( x,y +( k −α) D k ( x,y ))= y,
where x and y are row and column indices, respectively, k is a reference image position indicating a view from the multiple view data, and the placement matrix P k (x,y)=ŷ provides a pixel location (x′,ŷ) in the reference image that should be mapped on the pixel location (x′,y′) in a virtual view after warping; and
with the placement matrix, using a look-up table to conduct image warping.
2. The method of claim 1 , further comprising resolving any mapping of multiple pixels from different disparity levels onto the same pixel location by favoring pixels with larger disparity when generating foreground and for other occasions where there exist two pixels with column indices y m and y n , and row index x such that y m ≠y n and y′=y m −αD k (x,y m )=y n −αD k (x,y n ), using the following rule:
P
k
α
(
x
,
y
′
)
=
{
y
m
if
D
k
(
x
,
y
m
)
>
D
k
(
x
,
y
n
)
y
n
otherwise
.
(
2
)
3. The method of claim 2 , further comprising a refining step to choose final pixel values based upon the neighborhood of a pixel to choose a more likely value amount two or more candidate values.
4. The method of claim 1 , further comprising:
conducting coordinate alignment between multiple views; and
repairing holes in said multiple views individually with respect to a local background layer; and
generating final views using repaired information from the multiple views.
5. The method of claim 1 , further comprising:
generating a common background layer for multiple views;
repairing holes in the multiple views on the common background layer; and
generating final views using repaired information from the common background layer.