Media Compression - Scalable Continuous Media Streaming Systems - page 36

Input
Str eam

S1 + MV

VLC
Decoder

+

Q 1 −1

MC

IDCT

B1

A1

MV

+ +

MC
DCT

−

+ +

MV

AMVR

A2

S2

Q 2 −1

MC

+ +

B2

IDCT

+

Q 2

DCT

−

Trans. Method Used

S1 S2

Quantizer
Regulator

VLC
Encoder

Requantization A1 A2

Output
Stream

B1 B2 Spatial Downscaling

Figure 2.9 Logical design of the integrated transcoder

2.5.3 Implementation Issues
Figure 2.9 depicts the logical design of the integrated transcoder. The switches S1 and S2
control whether requantization or spatial downscaling is activated. If requantization is used,
switches S1 and S2 will be connected to A1 and A2 respectively, otherwise they will be
connected to B1 and B2 respectively to activate spatial downscaling. The following sections
present in more detail the algorithms used in the integrated transcoder.

2.5.3.1 Drift Compensation Loop

Both requantization and spatial downscaling introduce noises in the transcoded video frames.
As MPEG-1 makes use of predictive coding for B frames and P frames, these noises can
accumulate along frames in the same group of pictures (GOP), further degrading the visual
quality in the predicted frames. To prevent this problem, known as drift, we can feed back
the noises to the predictor to compensate for the prediction errors. This is known as drift
compensation.
For requantization, error is introduced during the requantization process where larger quan-
tization step sizes are used. To compensate for the quantization errors, we can dequantize
(Q 2 − 1 ) the quantized DCT coefficients and compute the quantization error by comparing to
the original DCT coefficients. The quantization error is then stored in the corresponding error