An Efficient Mode Decision Algorithm for Combined Scalable Video Coding - Advanced Concepts for Intelligent Vision Systems - page 63

BL complexity Te m p o r a l complexity

0 1 2 3 4 5 6 7 8

I/P I/P

B0

BL

B1

B1

EL

Temporal level 0

i/p i/p

Temporal level 1

b0

b1 Temporal level 2

b1

Temporal level 3

b2 b2 b2 b2

Fig. 6. The use of information for combined scalability structure

Temporal cor for the current MB can be calculated from the mode complexity
and the motion vector of the previous temporal level at the corresponding MB
position. The estimated Temporal cor is calculated by Eq. (4).
Temporal cor = Temporal complexity

MV l− 1
val

= Mode l− 1

complexity ×

search size ,

(4)

where l is the temporal level as shown in Fig. 6. Mode complexity indicates the
proposed mode number in Table 1. The value of search size represents the maxi-
mum length of search range. MV val is defined by MV values in the corresponding
MB of the previous temporal level, which is expressed as

MV val = avg ( | MV x | + | MV y | ) n , (5)

where n is the number of MV in the corresponding MB. The number of MV can
be various because partitioned ME is permitted in H.264/AVC. It is a integer
value by round up.
Quality cor ·

Spatial cor is defined as BL complexity , which represents the complex-
ity of texture in the base layer. For spatial scalability, the image in the base layer
and the image in the enhancement layer are very similar. However, the visual
quality between the base layer and the enhancement layer is extremely different,
which is affected by the different quantization scales. Therefore, BL complexity is
calculated by

Quality cor · Spatial cor = BL complexity

= scale f actor × BL Mode complexity

1
log 2 ( dif f QP ) ,

scale f actor = (6)

dif f QP = | QP BL − QP EL | , 2 ≤ dif f QP < 51 ,