Pulse or digital communications – Bandwidth reduction or expansion – Television or motion video signal
Reexamination Certificate
1998-12-29
2002-10-08
Le, Vu (Department: 2613)
Pulse or digital communications
Bandwidth reduction or expansion
Television or motion video signal
C375S240050, C375S240270
Reexamination Certificate
active
06463100
ABSTRACT:
BACKGROUND OF INVENTION
1. Field of the Invention
The present invention relates to a video encoder, and more particularly to a method for controlling the quantization. The present invention proposes a method of performing an adaptive quantization control in consideration of the characteristic of the input picture and the capacity of buffer in an MPEG2 encoder.
2. Background of the Related Art
Generally, digital TV provides a superior picture quality on the TV receiver for the viewers. As a result, a growing interest in digital TV broadcasting has cultivated many efforts to compress video data for transmission and reproduction. Typically, a moving picture expert group (MPEG2) is used as the algorithm to compress video signals. Having a high compression rate ranging from approximately {fraction (1/40)} to {fraction (1/60)}, the MPEG2 algorithm enables digital data transmission of high quality through general broadcasting channels to entertain viewers at home. The MPEG encoder classifies a picture into an Intra (I) frame, a Predictive (P) frame or a Bidirectional (B) frame and a decoder decodes the picture based on the type of the frame. Each frame is also divided into macro block (MB) units constituted by 16×16 pixels.
For purpose of the following illustration, some terms will first be defined as follows. The parameter for quantization of DCT coefficients in a MB is called a “quantizer_scale”, the parameter determined by the buffer in a bit rate control of TM5 (the name of a test model used in establishment of MPEG2 standards) is called an “offset Q
j
”, and the parameter obtained by multiplying the offset Q
j
with an activity measure in a macro block is called a quantization parameter “mquant
j
”.
The quantization parameter is clipped as an integer ranging from 1 to 31 and is sent to a decoder as a 5-bit header information. The quantizer scale for quantizing the DCT coefficients is substantially determined by a q_scale_type function at the MPEG2. In particular, the relationship quantizer_scale=g(mquant
j
) is satisfied and a function g(·) is determined by the q_scale_type. There are two types of q_scale_type in the MPEG2. If the q_scale_type is ‘0’, g(·) falls on a linear function, otherwise if q_scale_type is ‘1’, g(·) is a non-linear function.
A method for performing a bit rate control proposed in the MPEG2 TM5 will be described briefly.
FIG. 1
is a block diagram of a video encoder in the background art comprising an image memory
101
storing the original image in units of field or frame; a subtractor
102
obtaining a residual signal between the original image and a reconstructed image; and a DCT section
103
performing a DCT conversion of the residual signal. The value obtained by the DCT conversion is quantized at a quantizer
104
and the quantized signal is inverse-quantized at a inverse quantizer
105
prior to an inverse DCT (IDCT) conversion through an IDCT section
106
. The IDCT converted signal is added to a motion-compensated signal at an adder
107
and stored in a memory
108
. The video data stored in the memory
108
is subjected to motion estimation and compensation at a motion estimation/compensation (E/C) section
109
and sent to the subtractor
102
and the adder
107
.
A complexity calculator
110
calculates the spatial activity of the image stored in the image memory
101
. The spatial activity is generally a measure of the picture complexity or level detail of a macro block and will be further explained below. An encoder controller
111
is the bit rate controller and controls the quantization rate of the quantizer
104
in consideration of the calculated spatial activity and the capacity of a buffer
115
. The VLC
112
variable length codes the quantized data and the motion vector (MV) coding section codes the motion vector from the motion E/C section
109
. The VLC encoded data and the encoded MV are input to the buffer
115
and are transmitted to a decoder in the form of a bit stream data.
Particularly, the quantization is coarse for high spatial activity and less coarse for lower spatial activity. Thus, the spatial activity is utilized to control the bit rate control for quantization. Also, a defined bit rate is allocated to a group of pictures (GOP) according to a transfer bit rate and the bits are allocated to each picture according to the complexity of each picture I, P, and B. The global complexity X of each picture is given by Equation 1 below,
X
i
=S
i
Q
i
,X
p
=S
p
Q
p
,X
b
=S
b
Q
b
[Equation 1]
where S
i
, S
p
and S
b
are bits generated after the previous I, P and B pictures are encoded, and Q
i
, Q
p
and Q
b
are averages of the quantization parameters mquant
j
used in all macro blocks.
The complexity of the previous picture I, P, and B is used to obtain the bit allocation for the current picture of the same type and can be expressed by Equation 2 below.
T
i
=
max
{
R
1
+
N
p
X
p
X
i
K
p
+
N
b
X
b
X
i
K
b
,
bit_rate
8
×
picture_rate
}
T
p
=
max
{
R
N
p
+
N
b
K
p
X
b
X
p
K
b
,
bit_rate
8
×
picture_rate
}
T
b
=
max
{
R
N
b
+
N
p
K
b
X
p
X
b
K
p
,
bit_rate
8
×
picture_rate
}
[Equation 2]
In Equation 2, K
p
and K
b
are constants depending on the quantization matrix, typically having values 1.0 and 1.4 in the TM5, respectively. R is the number of bits remaining after encoding the previous picture bits allocated to the GOP. The bit-rate is a channel transfer rate (bits/sec) and the picture-rate is the number of pictures decoded per second.
The value of R is adjusted when the bits are allocated to the pictures in the next GOP as in Equation 3,
R←G+R
[Equation 3]
where G=bit_rate×N/image_rate and N is the size of GOP. N
p
and N
b
are the numbers of P and B images to be encoded within the current GOP.
The bit rate is controlled to encode the current picture at a rate adequate for the number of bits allocated, which is dependant upon the complexity in a picture. Also, assuming that an virtual buffer is assigned to each picture, the quantization parameters are regulated according to the state of the buffer. The state of each buffer may be expressed by Equation 4 before macro block j is encoded.
d
i
j
=d
i
0
+B
j−1
−{T
i
×(
j
−1)}/MB_cnt
d
p
j
=d
p
0
+B
j−1
−{T
p
×(
j
−1)}/MB_cnt
d
b
j
=d
b
0
+B
j−1
−{T
b
×(
j
−1)}/MB_cnt [Equation 4]
The values d
i
0
, d
p
0
and d
b
0
are the initial buffer values, which are actually the bit rate control difference of a picture from a previous picture of the same type. In other words, the initial buffer values are the differences between the number of bits allocated in coding the picture and the number of bits generated in coding a previous picture of the same type. MB_cnt is the total number of macro blocks for the image.
An offset Q
j
of the jth macro block is calculated by the following expression using the status information of the buffer when coding the (j−1)th macro block,
Q
j
={31
×d}/&tgr;
[Equation 5]
where r=2×bit_rate/picture_rate.
Adaptive quantization is a method of changing the offset Q
j
according to the complexity of the current macro block used to enhance the subjective quality of the picture. Particularly, Q
j
is multiplied by a factor N_act
j
utilizing an act
j
value indicating the complexity of macro blocks. The factor N_act
j
may be expressed by Equation 6 below,
N_act
j
=
2
×
act
j
+
avg_act
act
j
+
2
×
avg_act
[Equation 6]
where act
j
represents the minimum of the variances in the subblocks of the macro block. In Equation 6, the act
j
is smaller than the average complexity of a current picture for portions sensitive to the human's sight and accordingly the N_act
j
factor is also small. However, the N_act
j
Cho Sanghee
Lee Heesub
Lim Kyoung-won
Min Cheol-Hong
Fleshner & Kim LLP
Le Vu
LG Electronics Inc.
LandOfFree
Adaptive quantization control method does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Adaptive quantization control method, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Adaptive quantization control method will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2984197