An Efficient Mode Decision Algorithm for Combined Scalable Video Coding - Advanced Concepts for Intelligent Vision Systems - page 60

list 0 list 1

0 1 2 3 4 5 6 7 8

I/P I/P

B0

Low resolution
temporality
quality

BL

B1 B1

Inter layer
prediction

High resolution
temporality
quality

i/p i/p

b0

EL

b1

b1

b2 b2 b2 b2

Fig. 3. The structure of combined scalability

level is called the key picture and is an encoded as an intra (I-picture) or a
predictive frame (P-picture).

Combined Scalability. These types of scalability can be combined. The struc-
ture of combined scalability has features of both spatio-temporal and quality
scalability. For example, Fig. 3 shows the encoding structure for the combined
scalability with two layers. The base layer is encoded at a lower resolution, coarse
quality, and a slow temporal rate. The enhancement layer is encoded at a higher
resolution, fine quality, and a faster temporal rate.
The enhancement layer is also divided into an inter layer prediction picture
(i/p, b0, and b1) and a non-inter layer prediction picture (b2). To increase the
coding e ciency, the inter layer prediction picture can use the encoded infor-
mation of the base layer, such as intra texture, motion vector, and residual
coecients. The non-inter layer picture is only encoded by using bi-directionally
adjacent frames in the same layer.

2.2 Mode Decision in SVC

H.264/AVC uses the RD optimization (RDO) technique to determine the best
MB coding mode in terms of minimizing bit rates and maximizing image quality.

Inter Mode Prediction. Unlike the inter MB mode of previous video coding
standards, MB for H.264/AVC can be motion estimated and compensated from
already transmitted multi reference frames with varying block sizes from 16 × 16

down to 4 4 as shown in Fig. 4. One of these various types is determined as the

best inter mode. To determine the best inter MB mode in terms of minimizing
bit rate and maximizing image quality, the RDO technique is used. The best
inter MB mode is determined as one having the smallest RD cost in Eq. (1).

×

J inter = SSD { s, r ( MV ) } + λ motion · R ( MVD,REF ) , (1)