JPH04357787A - Picture encoding transfer rate control system - Google Patents

Abstract

PURPOSE:To accurately estimate the encoding information amount to decide an encoding parameter in the picture encoding transfer rate control system. CONSTITUTION:The frame encoding information amount is supplied as the input of an adaptive filter 106b, the tap coefficient of this adaptive filter 106b is updated for each frame period so as to reduce the error between the input and output of this adaptive filter 106b to the minimum, and the encoding parameter is decided according to the output of this adaptive filter 106. The output of the adaptive filter 106b shall be the estimation value of the present frame encoding information amount by the past frame encoding information amount weighted by the tap coefficient, thereby permitting accurate estimation and proper encoding parameter decision is implemented.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image encoding transmission rate control system.

0002

2. Description of the Related Art A moving image packet communication system using an asynchronous transfer mode network (ATM network) is attracting attention. The packet communication method transmits encoded audio and images into packets (
This system performs large-capacity communication by sending and receiving cells on an ATM network by grouping them into cells (cells), adding a header indicating the destination and type of content, and transmitting and receiving the cells over an ATM network. By taking advantage of the large capacity and asynchronous features and performing image encoding at a variable rate, high quality image communication becomes possible. In this case, if the total number of cells exceeds the processing capacity of the ATM network, the overflowing cells will be discarded. Regarding such problems, the following document describes a method for controlling the amount of encoded information (encoding rate) generated from a variable rate video encoder, taking into consideration network management, pricing, and service provision. It talks about. Literature title: Takishima, Wada "Rate control method in packet video coding", 1990 Image Coding Symposium (PCSJ90), pp. 239-242, 199
October 0

In this document, the amount of encoded information for one frame outputted from the encoder output terminal is monitored,
Step size Q of the quantizer of the frame In to be encoded
For example, n is determined as follows. ΔQ=a(In-1-Ifc) Qn=Qn-1+ΔQ ΔQ: Incremental value of Q a: Sensitivity coefficient In-1: Coded information amount If of the n-1th frame
: Short-term target information amount = [Io(n-1+N)-ΣIi]
/N, i=1~n-1 (average of n-1+N frames is I
Ifc: Correction value of If=kIo+(1-k)If(
k<0) Io: Target encoded information amount N for a frame
:Rate convergence parameter When quantization is performed with the step size Qn of the quantizer determined by this formula, the average encoding rate converges to Io.

0004

[Problem to be Solved by the Invention] However, in the above literature, when estimating the amount of encoded information of the frame to be encoded and determining the quantization step, a simple linear method of the amount of encoded information of the past frame is used. Since it uses a sum, it has the disadvantage that the estimation of the amount of information is not very accurate. Therefore, an object of the present invention is to provide a rate control method with high accuracy in estimating the amount of encoded information.

[0005]

[Means for Solving the Problems] The present invention has an adaptive filter, and provides the amount of encoded information for one frame or a predetermined unit as an input to the adaptive filter, and the output of the adaptive filter and the amount of encoded information for one frame or a predetermined unit. The magnitude relationship between the unit and the target encoded information amount is compared, and depending on the comparison result, the encoding parameter is changed stepwise in the direction of increasing or decreasing the frame encoded information amount. Then, for the next frame, the tap coefficients of this adaptive filter are sequentially updated every frame period in a direction that reduces the error between the input and output of the adaptive filter.

[0006]

[Operation] The adaptive filter outputs a value obtained by weighting the encoded information amount of past frames with tap coefficients and adding them. By sequentially updating the error between the input and output of the adaptive filter using a system identification algorithm such as the steepest descent method, the adaptive filter estimates the amount of encoded information at the time of the frame to be encoded. Output. This estimated amount of encoded information reflects changes in the amount of information of the past image itself and changes in past encoding parameters, and is compared with the target amount of encoded information, and depending on the result,
By changing the encoding parameters in stages for each frame, it can be expected to converge to the target encoded information amount.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing an embodiment of an encoding apparatus according to the present invention. In FIG. 1, the input original image is T
By scanning the image pickup tube of the V-camera 101, it is converted into an electrical signal indicating the density distribution of light, which is 8-bit quantized and sampled as a pixel by the A/D converter 102, and converted into a digital image signal. This digital image signal is
The image signal of each band is divided into four bands by an analysis filter 103 (QMF), encoded by encoders 104a to 104d, multiplexed by a multiplexer 105, and sent to a transmission path 107. . Note that the encoders 104a to 104d here are configured by DCT intraframe encoding, motion compensated interframe encoding+DCT, and the like. The rate control circuit 106 receives the encoded information from the multiplexer 105, calculates the amount of encoded information per frame in the code amount adder 106a, and applies the encoded information per frame to the adaptive filter 10.
6b calculates the estimated coded information amount of the current frame from the past coded information amount, and the step setting circuit 106c calculates the code of the current frame according to the comparison result between the estimated coded information amount and the target coded information amount. Determine the parameters.

The estimated encoded information amount 1wn, which is the output of the adaptive filter 106b, can be expressed by equation (1). 1wn=ΣAkn・In−k (k=1, 2, , K
) Equation (1) where, Akn: k-th tap coefficient at the n-th frame time of the adaptive filter;
In-k: encoded information amount of the n-kth frame, and K in this example is 5. The tap coefficients of the adaptive filter 106b are adjusted every frame period in the direction that the error between the input and output of the filter (that is, the estimation error of the encoded information amount) En shown in equation (2) becomes smaller for the next frame. Update. En=In-Iwn=In-ΣAkn・In-k(
k=0, 1, , K-1)

Equation (2) In this embodiment, the steepest descent method is used in which the square of the error En is used as the evaluation function, and updates are performed every frame period according to Equation (3). Akn+1=Akn+ΔA×sign(EnXIn
-k) (k=0, 1, , k-1)

Equation (3) where ΔA: increment of tap coefficient; sign: 0 if the value in parentheses is 0; -1 if negative; +1 if positive. In this way, by continuing to update the tap coefficients every frame period over several tens of frames, the adaptive filter 106b outputs an accurate estimated amount of encoded information along the change curve of the amount of encoded information in the past. It turns out.

The step setting circuit 106c determines the quantization step size according to the following equation and sets the quantization step size in the encoder 1.
40a to 40d to control the amount of encoded information. QBn=QBn-1+ΔQB (Iwn≧I
o) QBn=QBn-1-ΔQB (Iwn
<Io) Formula (3) However,
QBn: quantization step size of each band (QBmin<
QB<QBmax), ΔQB: step size increment set for each band in consideration of human visual characteristics, Io: target encoded information amount. In this way, the estimated amount of encoded information Iwn and the target amount of encoded information Io are compared, and depending on the comparison result, the encoding parameters are set in units to increase or decrease the amount of encoded information. Step (ΔQ
B) Change. By performing encoding using the encoding parameters thus determined, the amount of encoded information gradually converges to the target amount of encoded information in stages.

FIG. 2 is a block diagram of main parts of another encoding device according to the present invention, showing a portion corresponding to one of the four divided bands. Furthermore, this embodiment employs motion-compensated interframe predictive coding, sets a target amount of encoded information for each band, estimates the amount of encoded information for each band, and estimates the amount of encoded information for each band. The unit change amount of the encoding parameter is set for each band. In FIG. 2, first, an input image signal from an image signal input terminal 200 is sent to a frame memory 202 by a motion vector detector 201.
A motion vector is detected by matching with the image signal of one frame before, and an adder 203 subtracts a predicted signal that takes the motion vector into account to obtain a residual signal.

[0011] This residual signal is subjected to DCT transformation by a DCT transformer 204. The DCT coefficients are processed by the quantizer 20
5. quantized and dequantized by the dequantizer 206;
A variable length encoder 209 performs variable length encoding. The output transform coefficients of the inverse quantizer 206 are subjected to inverse DCT transform by an inverse DCT transformer 207, added with the previously subtracted prediction signal by an adder 208, and stored in the frame memory 202 as a locally reproduced image signal. The code of the coefficient and the motion vector are output from the code output terminal 210. Code amount adder 211, adaptive filter 212, and step setting circuit 2
13 estimates the encoded information amount of this band, compares it with the target encoded information amount of this band, and determines the quantization parameter of this band according to the result of the magnitude comparison. This quantization parameter is given to a quantizer 205 and an inverse quantizer 206 to control the amount of encoded information. This quantization parameter or its unit change is also output to the code output terminal 210.
is output from. Note that the quantization parameter determined here is a reference amount for this frame, and is also modified based on a preset instantaneous upper limit information amount, but a detailed explanation thereof will be omitted.

0012

Effects of the Invention As explained in detail above, in the present invention, the code amount of a frame to be encoded is compared to the code amount of past frames by using an adaptive filter whose tap coefficients are updated by a system identification algorithm. Since the estimation is performed using the linear sum of , the code amount estimation accuracy is high and appropriate rate control can be performed.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an encoding device according to the present invention.

[Figure 2]
Block diagram of another encoding device according to the present invention

[Explanation of symbols]

103 Analysis filter 104a-d Encoder 105 Multiplexer 106 Rate control circuit 106a Code amount adder 106b Adaptive filter 106c Step setting circuit

Claims (1)

[Claims] 1. An amount of encoded information for one frame or a predetermined unit is given as an input to an adaptive filter, and a magnitude relationship between the output of the adaptive filter and a target amount of encoded information for one frame or the predetermined unit is determined. and, depending on the comparison result, change the encoding parameter to increase or decrease the encoded information amount, and to reduce the absolute value error between the input and output of the adaptive filter. An image encoding transmission rate control method characterized in that the tap coefficients of the adaptive filter are updated sequentially at each frame period.