Compression - 3DTV: Processing and Transmission of 3D Video Signals - page 87

View 37 View 39 View 41

Unprocessed

Joint trilateral
filtering

Figure 3.14 Multi-view depth maps before and after tri-lateral filtering (Repro-
duced with the permission of Masayuki Tanimoto)

texture value, respectively:

coef f pq = c ( p , q ) · s depth ( p , q ) · s colour ( p , q ) (3.2)

The filters related to these three factors are considered as Gaussian filters
centred at point p . Accordingly, these individual factors are denoted as:

exp

c

2 p

q 2

1

2

c ( p , q ) = − − /σ

exp

2 d p −

d q 2

1

2
s depth

= − /σ

s depth ( p , q ) (3.3)

exp

2 I p −

I q 2

1

2
s colour

s colour ( p , q ) = − /σ

where, d represents the depth values of pixel points p and q ,and I represents
the corresponding colour texture luminance values. Figure 3.14 shows an
example snapshot from a test multi-view depth map sequence (three views
in a row), before and after the application of the colour-aided tri-lateral filter.
It is seen that the majority of textural gradients in the original depth maps
are smoothened, while the depth edge transitions are preserved. Reported
coding results have suggested that applying the processing filter on the
extracted multi-view depth map video sequences can reduce the bit-rate for
encoding the depth maps effectively by up to 52%.

3.4 Recent Trends in 3D Video Coding

In March 2011, MPEG released a new call for proposals for encoding generic
3D video signals, which can be considered a follow-up to the multi-view