Graphics Reference
In-Depth Information
6.1
Introduction
In the block-based hybrid video coding approach, transforms are applied to the
residual signal resulting from inter- or intra-picture prediction as shown in Fig.
6.1
.
At the encoder, the residual signal of a picture is divided into square blocks
of size
N
N
where
N
D
2
M
and
M
is an integer. Each residual block (
U
)is
then input to a two-dimensional
N
N
forward transform. The two-dimensional
transform can be implemented as a separable transform by applying an
N
-point one-
dimensional transform to each row and each column separately. The resulting
N
N
transform coefficients (
coeff
) are then subject to quantization (which is equivalent
to division by quantization step size
Qstep
and subsequent rounding) to obtain
quantized transform coefficients (
level
). At the decoder, the quantized transform
coefficients are then de-quantized (which is equivalent to multiplication by
Qstep
).
Finally, a two-dimensional
N
N
separable inverse transform is applied to the de-
quantized transform coefficients (
coeff
Q
) resulting in a residual block of quantized
samples which is then added to the intra- or inter-prediction samples to obtain the
reconstructed block.
Typically, the forward- and inverse transform matrices are transposes of each
other and are designed to achieve near lossless reconstruction of the input residual
block when concatenated without the intermediate quantization and de-quantization
steps.
a
b
Input block
Bitstream
Entropy
decode
Intra/Inter
prediction
U
level
Forward
transform
C
De-quant
Qstep
X
= coeff
Q
coeff
Inverse
transform
C
Qstep
Quant
level
Entropy
encode
Intra/Inter
prediction
Reconstructed
block
Bitstream
Fig. 6.1
Block-based hybrid video coding. (
a
) Encoder, (
b
) Decoder.
C
is the transform matrix
and
Qstep
is the quantization step size. Reproduced with permission from [
6
]. © IEEE 2013