JPH10155152A - Image compression coder and its method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image coder and its method by which a picture is coded at an object bit rate while keeping image quality of a reproduced image stably. SOLUTION: A rate control section 3 obtains an actual bit production quantity Bmb per one block every time each macro block of one picture is sequentially coded, obtains an accumulated bit production error Dmb by the coding so far and a quantization width q-scale is obtained by substituting the bit error Dmb to a function f (R, Dmb). The quantization width q-scale is given to a basic coding processing section 1. The basic coding processing section 1 applies quantization to each transformation coefficient obtained by DCT transformation in response to the quantization width q-scale, the quantization data are subject to variable length coding and compression coded image data are generated and the image data are sent through an encoder transmission buffer 2.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image compression encoding apparatus and method for compressing and encoding a video signal, and more particularly to control of a quantization width in encoding and control of a bit generation amount. It is about.

[0002]

2. Description of the Related Art As a technique relating to compression encoding of a video signal, for example, there is an international standard format of digital compression encoding data in ISO / IEC 13818-2 (commonly known as MPEG2), and a decoding method thereof is shown here. I have. Further, as a typical method of encoding in the format, there is a description in "Test Model 3" of ISO-IEC / JTC / SC29 / WG11 N0328.

In MPEG2 encoding, a video signal is subjected to motion compensation and prediction between pictures, and DCT encoding of a prediction error. When quantizing the transform coefficients of the DCT, the bit generation amount increases or decreases depending on the quantization width. When a video signal is compressed to a desired data amount, the quantization width is controlled according to the amount of bits generated for encoding, and the data amount is adjusted.

[0004] ISO-IEC / JTC / SC29 / WG11 N0328 "Test Mod
In el 3 ", a method for determining the quantization width of a video signal so that the video signal can be reproduced by inputting the encoded image data at a desired constant bit rate to a decoding device is shown.

In "Test Model 3", a target number of bits is converted from a target bit rate for each group of pictures (hereinafter referred to as a GOP) composed of a plurality of pictures, and the target number of bits is calculated for each picture I of the GOP. ,
P and B (picture I is at the head of the GOP, picture P is set for each of a plurality of pictures, picture B is between I and P), and these pictures are coded.

[0006]

However, in such a method, the target bit generation amount of the current GOP is adjusted and obtained based on the bit generation amount used for encoding each picture of the past GOP. In order to allocate the target bit generation amount to each picture of the GOP, a scene change, an increase in video motion in the picture, etc.
The ratio of the information amount of each picture I, P, B is the past GOP
In the case where the bit amount is significantly different from the above, the bit amount allocated to each of the pictures I, P, and B becomes inappropriate. The ratio of the information amount of each of the pictures I, P, and B often changes by a factor of two or more even in a normal case, and when there is a scene change, the same bit amount as that of the picture I is assigned to the picture B. If it becomes necessary or the scene becomes more complex, the amount of information may increase by a factor of nearly ten, and the amount of bits allocated to the picture will be insufficient, and the quality of the reproduced image corresponding to this picture will increase. There was a problem that it deteriorated.

Further, since the bit generation amount is controlled in GOP units, for example, when a scene change or the like occurs near the latter half of the GOP, that is, in the last few pictures of the GOP, the bit amount to be allocated is completely insufficient. Because
There is a problem that the degree of difficulty in encoding a picture sharply increases and the image quality of a reproduced image corresponding to the picture is extremely deteriorated.

On the other hand, in one picture, a target bit generation amount is calculated by assuming that each bit amount used for encoding each macroblock constituting the picture is constant. The quantization width of the video signal is controlled based on the comparison of the actual bit generation amount, and adjustment is performed so that the actual bit generation amount becomes the target bit generation amount in one picture.

[0009] However, the amount of information may be extremely different depending on the position of the video in the picture and its background, and the bit allocation in the picture cannot be performed well. For example, if a complex pattern exists in the second half of each macroblock of a picture, many bits will be allocated to the simple part of the first half, and the bit generation amount will be more than expected in the second half, or conversely If there is a complicated picture in the first half, the bit generation amount is suppressed more than necessary in the first half, and the bit generation amount is allocated more than necessary in the second half, and the part that should be allocated in one picture There was a problem that bits could not be allocated.

Further, even in the control method of adjusting the actual bit generation amount to the target bit generation amount, the occupation amount of the VBV buffer (virtual buffer) assumed on the decoding device side is not taken into consideration. Separately, it is necessary to correct the bit allocation and the like for each picture according to the constraint conditions from the VBV buffer, and in this case, the image quality of the reproduced image is further deteriorated.

In order to prevent the image quality of a reproduced image from becoming unstable and deteriorating due to such a change in the degree of difficulty in encoding the video signal, the quantization width of the video signal should be kept as uniform as possible. However, if the quantization width is fixed, the amount of bits generated randomly increases in accordance with the degree of difficulty in encoding an image, and it becomes difficult to encode a video signal at a target bit rate.

The present invention solves the above-mentioned conventional problem. An appropriate bit amount can always be assigned to a picture at the time of a scene change or a complicated picture. The optimal bits can be assigned to each macroblock according to the complexity of each macroblock, and the picture can be encoded at the target bit rate while maintaining the image quality of the reproduced image stably. An image compression encoding apparatus and method are provided.

[0013]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention obtains a quantization width for encoding a video signal, and calculates an image for encoding the video signal based on the quantization width. A compression encoding apparatus, wherein an actual bit generation amount used for encoding the video signal is obtained, a target bit generation amount corresponding to a target bit rate is obtained, and the actual bit generation amount and the target bit generation amount are calculated. A bit generation error, which is a difference between the bit generation amounts, is obtained, and a quantization width corresponding to the bit generation error is obtained.In a predetermined range of the bit generation error, a change in the bit generation error is determined. There is provided control means for setting a characteristic in which the quantization width changes small, and for obtaining a quantization width for a bit generation error based on the characteristic.

In one embodiment, a reference quantization width is predetermined, and the control means calculates a ratio of a value obtained by subtracting a value corresponding to a bit generation error from a target bit rate and the target bit rate. The ratio and the reference quantization width are integrated, and the quantization width when encoding the video signal is determined based on the product.

In one embodiment, the control means obtains a reference quantization width based on the target bit rate, and in a predetermined range of the bit generation error, the control unit determines the quantization width with respect to a change in the bit generation error. A characteristic that changes slightly near the reference quantization width is set, and a quantization width for a bit generation error is determined based on this characteristic.

In one embodiment, in a predetermined range of the bit generation error and in the vicinity of the predetermined range, the characteristic of the change of the quantization width with respect to the change of the bit generation error draws a hysteresis loop.

In one embodiment, when a virtual buffer for sequentially inputting image data obtained by encoding a video signal and sequentially outputting the image data is assumed, the occupation amount of the virtual buffer is determined by the number of bits generated. In response to the error, when the occupation amount of the virtual buffer becomes close to 0, the quantization width sharply approaches the maximum value, and is connected to the characteristic of the quantization width that changes little with the change of the bit generation error. , The quantization width changes continuously.

In one embodiment, a plurality of ranges of the bit generation error are predetermined, and a characteristic indicating a change in the quantization width with respect to the change in the bit generation error is set for each of these ranges. From the range, select the range in which the bit generation error falls, and based on the characteristics of this range,
The quantization width for the bit generation error is determined.

In one embodiment, the video signal is encoded in macroblock units based on the quantization width in accordance with the MPEG standard, and the control means controls, for each macroblock, a quantum signal corresponding to a bit generation error. Seeking the extent of change.

Next, the operation of the present invention will be described.

According to the present invention, a bit generation error which is a difference between an actual bit generation amount and the target bit generation amount is obtained, and a quantization width corresponding to the bit generation error is obtained. In the predetermined range, a characteristic is set in which the quantization width changes small with a change in the bit generation error, and the quantization width for the bit generation error is obtained based on this characteristic. For this reason, in a state where the bit generation error changes within a predetermined range of the bit generation error, the quantization width does not change significantly,
Approximately the same target bit generation amount is assigned to each of the pictures I, P, and B in the GOP, and the image quality of each reproduced image corresponding to these pictures is stabilized.
Also, in one picture, as long as the bit generation error changes within a predetermined range of the bit generation error, an appropriate quantization width is set, and an appropriate quantization width is set according to the difficulty of encoding the picture. It is possible to assign a desired target bit generation amount.

As in one embodiment, a ratio between a value obtained by subtracting a value corresponding to a bit generation error from a target bit rate and the target bit rate is obtained, and this ratio and the reference quantization width are integrated. The quantization width when encoding the video signal may be obtained based on the product. In this case, as the bit generation error increases, the quantization width increases, and as the bit generation error decreases, the quantization width decreases.

As in one embodiment, within a predetermined range of the bit generation error, a characteristic is set in which the quantization width changes slightly near the reference quantization width with respect to the change of the bit generation error. The quantization width for the bit generation error may be obtained based on the characteristics. Thus, an appropriate quantization width can be set in the above range.

As in one embodiment, in the predetermined range of the bit generation error and in the vicinity of the predetermined range, the characteristic of the change of the quantization width with respect to the change of the bit generation error may form a hysteresis loop. In this case, if the bit generation error changes until it goes out of the range and the quantization width changes accordingly, even if the bit generation error returns to the range, the quantization width does not return to the same value. , The characteristic of the change of the quantization width with respect to the change of the bit generation error becomes new.

As in one embodiment, the virtual buffer (V
(BV buffer) corresponds to the bit generation error, and when the occupation amount of this virtual buffer approaches 0, the quantization width sharply approaches the maximum value and changes little with the change of the bit generation error. The quantization width may change continuously so as to be connected to the characteristics of the quantization width. In this case, when the occupation amount of the virtual buffer is close to 0, the quantization width sharply approaches the maximum value, so that underflow of the virtual buffer can be prevented.

As in one embodiment, a range in which a bit generation error is included is selected from each range of the bit generation error, and the quantization width for the bit generation error is obtained based on the characteristics of this range. good. In this case, an appropriate characteristic can be selected according to the magnitude of the bit generation error, and the quantization width can be determined based on this characteristic.

As in one embodiment, the video signal may be encoded in macroblock units based on the quantization width in accordance with the MPEG standard. As described above, even in such a coding in units of macroblocks, an appropriate quantization width is set as long as the bit generation error changes within a predetermined range of the bit generation error. Thus, an appropriate target bit generation amount can be assigned according to the degree of difficulty in encoding a picture.

[0028]

Embodiments of the present invention will be described below with reference to the accompanying drawings. FIG. 1 is a block diagram schematically showing a first embodiment of the image compression encoding apparatus of the present invention. In the encoding apparatus according to the first embodiment, a video signal is encoded into image data by a basic encoding processing unit 1, and the image data is transmitted through an encoder transmission buffer 2. Further, the rate control unit 3 controls the quantization width q_scale when encoding the video signal by the basic encoding processing unit 1.

Here, the outline of the encoding process of the video signal will be described with reference to FIG. Each picture of the video signal to be encoded is divided into a plurality of macroblocks and encoded. One macroblock has a luminance of 1
Data of a 6 × 16 pixel block (further divided into four 8 × 8 pixel blocks, which are basic encoding processing units), and color differences are divided into data of two 8 × 8 pixel blocks, which are basic encoding processing units. , These blocks are encoded. Also, for one macroblock, the coding method and quantization width of the blocks in the macroblock are determined.

A slice is a data unit including a plurality of macroblocks, and one picture is composed of a plurality of slices. As for the picture encoding method, a picture (intra picture: hereinafter referred to as picture I) in which the picture itself is encoded as is, and a picture to be encoded after predicting motion from a temporally past picture (hereinafter referred to as picture I) P) There is a picture (hereinafter referred to as picture B) that is coded after predicting the motion from both or either temporally past and future pictures.

The arrangement of the pictures I, P, and B in FIG. 2 is a typical example. The picture P is predicted and encoded three pictures ahead using the first picture I, and each picture included in the picture P is predicted. Picture B is encoded by predicting it from both pictures I and P. Therefore, picture I is encoded first, picture P is encoded next, and each picture B is encoded. Therefore, the order of each picture I, P, B along the original time progression is determined. Must be changed before encoding these pictures.

Further, a GOP (group of pictures) is composed of a plurality of pictures starting from picture I, and one video sequence is composed of an arbitrary number of GOPs.

Such encoding is performed by the basic encoding processing unit 1 in the apparatus shown in FIG. This basic encoding processing unit 1
3 is shown in FIG.

In FIG. 3, reference numeral 11 denotes an image rearranging unit for inputting a video signal indicating each picture, and rearranging these pictures in an encoding order; and 12, a macroblock which is a unit for encoding the picture. A scan conversion processing unit 13 for converting the difference between the macroblock and the prediction data according to the video operation in the picture; and a DCT transform 14 for encoding the difference output from the difference unit 14 The processing unit 15 is a weighted quantization processing unit,
6 is a variable length coding processing unit, 17 is an inverse quantization unit, 18
Is an inverse DCT processing unit, 19 is a prediction processing unit with motion compensation, 20 is a mode determination unit, 21 is a motion detection processing unit,
22 is an addition processing unit.

In the encoding processing unit 1 having such a configuration,
The image rearranging unit 11 inputs each picture indicated by the video signal, rearranges the order of these pictures to the order of encoding, and sequentially outputs these pictures to the scanning line conversion processing unit 12.

The scanning line conversion processing section 12 divides a picture into a plurality of macroblocks each time each picture is input, and sequentially outputs these macroblocks to the difference section 13. The difference unit 13 obtains a difference between the macroblock from the scanning line conversion processing unit 12 and the prediction data from the prediction processing unit with motion compensation 19, and outputs this difference as a prediction error. The DCT transform processing unit 14 performs a DCT transform on the prediction error in units of 8 × 8 pixel blocks, and outputs each transform coefficient obtained by the DCT transform to the weighting quantization unit 15. The weighting quantization processing unit 15 quantizes each transform coefficient, and outputs the resulting quantized data to the variable-length coding processing unit 16. The variable-length encoding unit 16 performs variable-length encoding on the quantized data to form compression-encoded image data. The compression-encoded image data is temporarily stored in an encoder transmission buffer for transmission at a desired transmission rate, and then output.

On the other hand, the quantized data output from the weight quantization processing unit 15 is reproduced by an inverse quantization unit 17 and an inverse DCT processing unit 18 and then input to a motion-compensated prediction unit 19. To form prediction data indicating a picture whose motion has been predicted. This forecast data
The difference is input to the difference unit 13, where the difference between the prediction data and the macroblock from the scanning line conversion processing unit 12 is calculated.
The motion detection processing unit 21 calculates a motion vector of a video for each macroblock, inputs the motion vector to the prediction processing unit 19 with motion compensation, and sends the motion vector to the variable length coding processing unit 16. The rate control unit 3 compares the bit generation amount of the bit stream output from the variable length coding processing unit 16 with the target bit generation amount converted from the target bit rate, and finally completes the encoding with the target bit generation amount. The quantization width q_scale of the weighting quantization unit 15 is controlled so that

However, in the MPEG standard, VB
Assuming a virtual decoder called V, the input buffer of this VBV (hereinafter referred to as VBV buffer) must be controlled so as not to overflow or underflow. The input / output of this VBV buffer is defined.

FIG. 4 shows the data occupancy in the VBV buffer when encoding a video signal at a fixed rate. In FIG. 4, the horizontal axis represents time, and one scale is 1
Shows the time spent on picture input. The vertical axis is the data occupancy VBV_fullness of the VBV buffer. At first, the image data is not reproduced, and the image data is stored in the VBV buffer for a period VBV_delay indicated by the picture header of the image data, that is, a period from time T1 to time T2 when image data obtained by encoding one picture is input. To save. At this time, the occupation amount VBV_fullness of the VBV buffer is represented by a straight line with a constant slope because image data is input at a constant rate. next,
At time T3 (= T2) when the picture is reproduced, 1
Image data for a picture is instantaneously removed from the VBV buffer. By the time T4 immediately before the reproduction of the next picture is performed, the occupancy VBV_fullness of the VBV buffer is increased by the bit amount of one picture,
At time T5 (= T4) of picture reproduction, 1
Image data for the picture is removed. Thereafter, the encoder operates in a similar manner to control the transmission of image data so that the occupation amount VBV_fullness of the VBV buffer does not exceed the size VBV_size of the VBV buffer, that is, so that underflow and overflow do not occur. There is.

The variable-length encoding unit 16 of the basic encoding unit 1 sends an MB end timing signal to the rate control unit 3 every time the encoding of each macroblock is completed, and simultaneously transmits one picture. Each time encoding is completed, a picture start timing signal is sent to the rate control unit 3. Each time the MB end timing signal is input, the rate control unit 3 obtains the actual bit generation amount Bmb per macroblock based on the bit stream from the variable length coding processing unit 16, and the bit generation amount Bmb The quantization width q_scale is determined based on this, and the quantization width q_scale is sent to the basic encoding processing unit 1.

FIG. 5 shows the configuration of the rate control unit 3. In the figure, the MB counter 4 counts the amount of bits generated used for coding each macroblock based on the bit stream from the variable length coding processing unit 16. The control circuit 5 receives the MB end timing signal and the picture start timing signal, and obtains the actual bit generation amount Bmb per macroblock based on the count value of the MB counter 4 every time the MB end timing signal is input. . Further, the control circuit 5 receives the target bit rate and the size VBV_size of the VBV buffer from outside, and based on the actual bit generation amount Bmb, the target bit rate, the size VBV_size of the VBV buffer, and the like,
A quantization width q_scale is obtained for each macroblock, and the quantization width q_scale is given to the weighting quantization unit processing unit 15 of the basic coding processing unit 1. Further, the control circuit 5 controls the VBV
When the buffer approaches an underflow state, an underflow avoidance signal is output to the weighting / quantization unit processing unit 15. Further, the control circuit 5 outputs to the variable encoding processing unit 16 a signal indicating the VBV_delay period, the target bit rate, and the VBV buffer size VBV_size as the header information of the image data.

Such processing is performed according to the flowchart of FIG. First, the control circuit 5 of the rate control unit 3
Is the target bit generation amount Tmb per macroblock,
Target bit generation amount Tpic per picture, target bit rate R (bps), reference quantization width q_st, bit generation error Dmb, VBV buffer occupancy VBV_fullnes
s, VBV buffer size VBV_size, quantization width q_s
Initialize cale (step 101).

In this initialization, assuming that the target bit rate is externally received as the rate per macroblock, the target bit rate is determined as the target bit generation amount Tmb per macroblock. The target bit generation amount Tpic per picture is obtained by multiplying the initial value of the target bit generation amount Tmb by the number of macroblocks in one picture by an integer (Tpic = Tmb ×
(Number of macroblocks per picture)). The target bit rate R (bps) is obtained by multiplying the target bit generation amount Tpic per picture by the frame rate (R = Tpic × (frame rate)).

Assuming that encoding is performed using a default quantization matrix defined by the standard, when the target bit rate R (bps) is approximately 3 Mbps, the reference quantization width q_st is set to 6, Target bit rate R (bp
If s) is 4 Mbps or more, the reference quantization width q_st is set to 5, and the target bit rate R (bps) is 6 Mbps.
If it is not less than s, the reference quantization width q_st is set to 4.
As described above, the reference quantization width q_st is set to the target bit rate R (b
ps), but this reference quantization width q_s
t may be appropriately changed and set.

Here, the target bit rate R (bps)
Is set to 3 Mbps, and the reference quantization width q_st is set to 6.

The initial value of the bit generation error Dmb is set to 0. The size of the VBV buffer is specified externally,
For example, 1835008 specified by MP @ ML standard
Is a bit. VBV buffer occupancy VBV_fullnes
s is the bit generation error Dm as is clear from the graph of FIG.
b, and the initial value of the occupancy VBV_fullness corresponds to the initial value 0 of the bit generation error Dmb 13350
Set to 08 bits. Therefore, when the position of the initial value 0 of the bit generation error Dmb on the horizontal axis of the graph of FIG. 8 is changed, the initial value of the occupation amount VBV_fullness is also changed. The initial value of the quantization width q_scale is the reference quantization width q_st
Is set to 6, which is the same as

When the initialization is completed, the control circuit 5
Determines whether or not the end of a series of video signals is based on the output of the variable length coding processing unit 16 (step 10).
2) Subsequently, it is determined whether or not it is the start of a picture (step 103). Here, since the encoding of the video signal has just started, it is determined that the end of a series of video signals has not been completed (step 102, No), and it is determined that the picture has started (step 103, Yes). Move on to stuffing processing.

The stuffing process in step 104 is performed according to the procedure shown in FIG. The control circuit 5 initializes the stuffing byte to 0 (step 20).
1), VBV buffer occupancy VBV_fullness and VB
The value obtained by subtracting the target bit generation amount per picture Tpic from the V buffer size VBV_size is compared (step 202), and the VBV buffer occupation amount VBV_f is compared.
If the ullness is greater (Step 202, Yes), the VBV
Since the buffer overflows, the stuffing byte is incremented and set to 1. Then, with the increase of the stuffing bytes, the occupation amount VBV of the VBV buffer
_fullness minus 8 to get this occupancy VBV_full
In addition to updating the ness, the bit generation error Dmb is updated by adding 8 to the bit generation error Dmb (step 20).
3) Then, the process returns to step 202.

Therefore, the occupancy VB of the VBV buffer
As long as it is determined that V_fullness is larger (Step 202, Yes), the stuffing byte is increased by one, and the occupancy VBV_fullnes of the VBV buffer is accordingly increased.
s decreases by 8, and the bit generation error Dmb increases by 8 (step 203). Then, when it is determined that the occupation amount VBV_fullness of the VBV buffer is smaller (Step 202, No), the value of the stuffing byte at that time is notified to the variable length coding processing unit 16. The variable-length coding processing unit 16 determines the value of the stuffing byte
The stuffing process, that is, sending a code string of 0 in byte units subsequent to the image data obtained by encoding the picture, apparently increasing the image data of this picture, and increasing the amount of image data removed from the VBV buffer; This avoids overflow of the VBV buffer.

However, in the initial state, the occupation amount VBV_fullness of the VBV buffer is determined to be smaller (step 202, No), so that the value 0 of the stuffing byte is notified to the variable length coding processing unit 16, and the stuffing is performed. No action is taken.

Thereafter, the control circuit 5 controls the period VBV_delay
Is calculated from the following equation (1) or (2) (step 10).
5). For pictures B and P, VBV_delay = 90,000 × (VBV_fullness + Tpic) / R (1) For picture I, VBV_delay = 90,000 × (VBV_fullness + Tpic−number of header bits) / R (2) where R is a target per second. The bit generation amount and the number of header bits are the number of bits from the video sequence header to immediately before the picture header.

This period VBV_delay is transmitted to the decoding device as header information of image data obtained by encoding a picture. In this decoding device, if image data is input to the VBV buffer for the first time, the decoding of image data is started after the image data is accumulated in the VBV buffer after waiting for a period VBV_delay included in the header information. .

During the processing so far, the encoding of one picture by the basic encoding processing unit 1 is started, and the control circuit 5 of the rate control unit 3 counts up by the MB counter 4 every time the MB end timing signal. The actual bit generation amount Bmb per macroblock is obtained based on the numerical value. (Step 106). The control circuit 5 calculates the actual bit generation amount Bmb per macroblock based on the following equation (3).
Is added to the bit generation error Dmb to update the bit generation error Dmb (step 1).
07). Dmb = Dmb + Bmb-Tmb (3) Here, if the information amount per macroblock is too large or too small, the actual bit generation amount Bmb becomes much smaller than the target bit generation amount Tmb, and the bit generation is performed. An error Dmb occurs. If the amount of information per macroblock is just right, the actual bit generation amount Bmb matches the target bit generation amount Tmb, and the bit generation error Dmb is 0.
Becomes

Subsequently, after confirming that the bit generation error Dmb does not exceed the predetermined threshold value D4 (step 108, No), the control circuit 5 determines the following equation (4).
A function f (R, R) based on (5), (6), (7), (8)
Dmb) to obtain a quantization width q_scale (step 11).
0). If Dmb <D0, f (R, Dmb) = 1 (4) If D0 ≦ Dmb <D1, f (R, Dmb) = q_st × R / (R−K1 × Dmb) (5) D1 ≦ Dmb < In the case of D2 f (R, Dmb) = q_st × R / (R−K2 × Dmb) (6) When D2 ≦ Dmb <D3 f (R, Dmb) = q_st × R / (R−K3 × Dmb) −K4 (7) When Dmb ≧ D3 f (R, Dmb) = 31 (8) where the above equations (4), (5), (6), (7), (8)
As is clear from the graph of FIG. 8 showing the characteristic of the function f (R, Dmb) based on D0 <D1 <D2 <D3,
= -400000, D1 = 0, D2 = 600000, D3
= 120000000. K1 is set to a value such that f (R, Dmb) = 1 when Dmb = D0, and K2
Is set to a value such that f (R, Dmb) = 9 around Dmb = 2, and K3 and K4 are set to f (R, Dm when Dmb = D2.
b) = 9, and f (R, Dmb) = around Dmb = D3
The value is set to a value close to 31. Here, K
1 = 37.5, K2 = 1.67, K3 = 2.0, K4 = 1.
Set to 0. Thereby, the characteristic curve F of f (R, Dmb) is obtained.
Changes continuously.

Thus, the quantization width q_scale = f (R, Dm
After obtaining b), the control circuit 5 rounds off the decimal point of the quantization width q_scale and then gives the quantization width q_scale to the weighting quantization processing unit 15 (step 11).
1). The weighting quantization processing unit 15 calculates the quantization width q_scale
, And quantizes each transform coefficient obtained by the DCT transform, and outputs the quantized data thus obtained to the variable-length coding processing unit 16. The variable-length encoding unit 16 performs variable-length encoding on the quantized data to form compression-encoded image data.

Thereafter, similarly, the processing of steps 102 to 111 is repeated, each macroblock is sequentially encoded, and each time, the actual bit generation amount Bmb per macroblock is obtained, A bit generation error Dmb is calculated, and this bit error Dmb is
(R, Dmb) to obtain a quantization width q_scale, quantize each transform coefficient obtained by DCT according to the quantization width q_scale, perform variable length coding on the quantized data, and perform compression coding. Forming the image data.

If the bit generation error Dmb exceeds the threshold value D4 while repeating the processing of steps 102 to 111 (step 108, Yes), the actual bit generation amount Bmb is larger than the target bit generation amount Tmb. If each bit is excessively used to encode a picture and the bits are insufficient, the VBV buffer may underflow. It is output to the conversion unit processing unit 15 (step 109). In response to this, the weighting quantization unit processing unit 15 encodes only the DC component of the macroblock, and reduces the code amount of the image data. This alleviates the overuse of each bit, and the actual bit generation amount Bmb approaches the target bit generation amount Tmb.

Here, as is apparent from the graph of FIG.
When D0 ≦ Dmb <D1, that is, the bit generation error Dmb
Changes in the vicinity of approximately 1/2 of the VBV buffer size VBV_size, the change in the quantization width q_scale is small. For example, when the quantization width q_scale is around 6,
Even if the bit generation error Dmb increases or decreases by about 300 Kbit, the change of the quantization width q_scale is about 1.

Therefore, as long as the bit generation error Dmb changes in the vicinity thereof, the quantization width q_scale is kept substantially constant, and the actual bit generation amount Bmb per macroblock is obtained.
Increase or decrease depends on the amount of information of the macroblock. For this reason, if the reference quantization width q_st (= 6) is appropriately set, even if the information amount increases or decreases in the latter half and the former half of each macroblock of one picture, the optimum bit generation for each macroblock is performed. Thus, the amount of bits generated for encoding one picture can be prevented from being insufficient or surplus.

Since the amount of bit generation consumed for encoding one picture also depends on the amount of information of the picture, the reference quantization width q_st (= 6) can be set appropriately. , The optimal bit generation amount can be given to each picture.

In a region where the quantization width q_scale is a relatively large value, the influence of the change in the quantization width q_scale on the amount of bits generated for encoding one picture is small, so that a relatively complex image can be reproduced. Even if the picture is coded, the target bit rate can be maintained.

Also, since the quantization width q_scale is determined based on the bit generation error Dmb, the same quantization width as that used when encoding the picture I is used when encoding the picture immediately after the scene change. With q_scale, encoding can be performed without greatly changing the quantization width q_scale. Even immediately after a scene change, a video signal can be encoded so that deterioration in the image quality of a reproduced image is reduced.

As is apparent from the above description, in the above-described embodiment, each time a macroblock is coded, the target bit generation amount converted from the bit generation amount used for the previous coding and the target bit rate is converted. A bit generation error Dmb, which is a difference between the amounts, is calculated, and a quantization width q_scale is determined based on a ratio between a value obtained by subtracting a value proportional to the bit generation error Dmb from the target bit rate and the target bit rate.

When encoding an image, it is said that for the same image, the product of the quantization width q_scale and the amount of bits generated for encoding becomes substantially constant, and the quantization width q_s
When the function f (R, Dmb) for deriving the cale is set, the amount of each bit generated for sequentially encoding substantially the same video is quantized width q_scale so that the above product is substantially constant. , And is set to a value obtained by subtracting a constant multiple of the bit generation error Dmb from the target bit generation amount. Therefore, the bit generation amount consumed for encoding can be controlled to the target bit generation amount in the same manner as controlling the bit generation amount consumed for encoding by the primary feedback control.

When D0 ≦ Dmb <D1, the coefficient K1 is set small so that the quantization width q_scale changes in the vicinity of 6 to 9. Therefore, in the range of this quantization width q_scale 6 to 9, , The amount of bits generated for encoding fluctuates,
The bit generation error Dmb easily increases and decreases, and the size VBV_size of the VBV buffer is effectively used. by this,
Until the capacity of the decoding apparatus is fully utilized, the amount of bits generated for encoding can be increased in accordance with an increase in the amount of information due to scene changes or severe video motion.

Conventionally, a target bit generation amount is set for each picture, and each time a macroblock in a picture is coded, control is performed such that the bit generation amount used for coding the macroblock approaches a constant value. The distribution of the complexity of the picture in the picture,
The quantization width of a macroblock immediately after coding of a simple picture became smaller, and the quantization width of a macroblock immediately after coding of a complicated picture became larger, and the image quality of the reproduced image tended to be uneven. .

On the other hand, in the present embodiment, control is performed such that the amount of bits generated for encoding varies by effectively utilizing the size of the VBV buffer having a much larger capacity than the picture. Therefore, it is possible to prevent the image quality of the reproduced image from becoming uneven due to the distribution of the complexity of the image in the picture.

In this embodiment, the control of the target bit rate and the control of the VBV buffer are described using both the bit generation error Dmb and the occupation amount VBV_fullness of the VBV buffer. Immediately after encoding a picture, the following equation (9) holds, and one can guide the other. VBV_fullness = (initial value of VBV_fullness) −Dmb (9) Therefore, if the equation (9) is used, the bit generation error Dmb
Only by the above, all processes, that is, a stuffing process for avoiding overflow (step 104),
Underflow avoidance processing (step 109) and the like are possible, and the present invention includes this. Similarly, all processing can be performed only by VBV_fullness.

In other words, conventionally, it was necessary to separately control the target bit rate and the VBV buffer. In the present invention, however, one parameter of the bit generation error Dmb and the occupancy VBV_fullness is used. It enables unified control by only

In the present embodiment, the size VBV_size of the VBV buffer is defined by the MP @ ML standard of MPEG2. However, the size VBV_size of the VBV buffer may be changed. A function for deriving the quantization width q_scale may be determined so that the occupation amount VBV_fullness is controlled.

The target bit rate R (bps) is set to 3
Although Mbps is set, the same method as in the present embodiment can be applied to other bit rates.

Further, the MPEG2 encoding apparatus is illustrated, but the present invention can be applied to an encoding apparatus and an encoding method similar to this.

In the present embodiment, when D0 ≦ Dmb <D1, the coefficient K1 is set small so that the quantization width q_scale changes in the vicinity of 6 to 9, but the code of the expected video signal is set. Is difficult to quantify (for example, in the case of a movie, etc., the encoding is relatively easy, and the quantization width q_scale
Can be set smaller), and if the image quality required for the reproduced image is low, the quantization width q_s near the threshold value D2
The cale may be set to about 8. For example, when the target bit rate is about 4 Mbps, the change of the quantization width q_scale in the range of D1 to D2 may be set to about 5 to 7, and the default quantization matrix defined by the standard may be set. Alternatively, when using a value close to the quantization matrix, the quantization width q_scale is set to 1 in the range of D1 to D2.
What is necessary is just to set it so that it does not exceed 0. Further, each of the threshold values D0, D1, D2, D3 and D4 can be shifted within a range of about several hundred Kbit. In short, in the present invention, the quantization width q_scale and the threshold values D
The values are not limited to 0, D1, D2, D3, and D4, and these values can be appropriately changed.

FIG. 9 shows the characteristics of another function f (R, Dmb) for deriving the quantization width q_scale. However, the reference quantization width q_st is set to 5.

The characteristics of the function f (R, Dmb) shown in FIG. 9 are represented by the following equations (10), (11), (12), and (13).

When Dmb <D0, f (R, Dmb) = 1 (10) When D0 ≦ Dmb <D1, f (R, Dmb) = q_st × R / (R−K1 × Dmb) (11) D1 ≦ Dmb <D2 f (R, Dmb) = q_st × R / (R−K2 × Dmb) (12) If Dmb ≧ D2, f (R, Dmb) = 31 (13) K1 is Dmb = F (R, Dmb) when -1,000,000
= 1, and K2 is Dmb = 6000.
At 00, f (R, Dmb) = 31 is set.

According to the characteristics shown in FIG. 9, the setting is highly effective for a series of pictures mainly including a still or small-moving image and a series of pictures in which scene changes frequently occur.

When a conventional encoding method is applied to such a series of pictures, the amount of bits generated for encoding is extremely concentrated on picture I, and the allocation of the bits is reduced for other pictures. And V
The occupancy VBV_fullness of the BV buffer increases, and stuffing processing tends to occur.

For this reason, here, it is mainly assumed that encoding is performed with a quantization width q_scale corresponding to the range from D0 to D1. Therefore, picture I is encoded with a quantization width q_scale corresponding to the vicinity of D0 in a video close to static,
Even if the amount of bits generated for encoding becomes very large, a low quantization width q_scale can be maintained. Even if a scene change occurs in the middle of a GOP, the quantization width q_scale is in the range from D0 to D1.
Is small, each bit is allocated according to the complexity of the picture.

If there is a tendency in the picture to be encoded in this way, the thresholds D0, D1, D2 and the coefficients K1, K2 are appropriately set, and a function f suitable for the tendency is set.
It is effective to determine the characteristics of (R, Dmb).

The minimum value of the quantization width q_scale is set to 1. However, when importance is placed on the uniformity of the reproduced image, the minimum value of the quantization width q_scale is set to 2 or more, and the quantization width q_scale is set.
It is effective to set the fluctuation of the small.

Also, a function f for deriving the quantization width q_scale
In (R, Dmb), a number of threshold values D0, D1,.
May be set to finely divide each range between the threshold values, and the coefficients K1, K2,... May be appropriately set for each of these ranges to derive the quantization width q_scale. Further, when each range between the threshold values is finely divided, the characteristic of the quantization width q_scale is approximated to a straight line for each range, and the quantization width q_sca
The calculation amount required for calculating le may be reduced. Also, the magnitude of each threshold may be changed according to the permissible image quality of the reproduced image.

FIG. 10 is a flowchart showing an encoding process according to the second embodiment of the present invention. This encoding process is performed in the encoding processing device shown in FIGS. 1, 3, and 5.

The processing of the flowchart of FIG. 10 is performed by the steps 101 and 1 in the processing of the flowchart of FIG.
10 is replaced with steps 101A and 110A, and the other steps 102 to 109 and 111 are common to the processing in the flowchart of FIG.

That is, the processing of the flowchart of FIG. 10 is different from the processing of the flowchart of FIG. 6 only in the initialization processing of each parameter in step 101A and the derivation processing of the quantization width q_scale in step 110A. This step 101A will be described next with reference to the flowchart of FIG.

First, the control circuit 5 of the rate control section 3 sets the target bit rate R (bps) to a predetermined value, sets the size of the VBV buffer to 1835008 bits, and sets the bit generation error Dmb to 0. Set and occupancy VB
The V_fullness is set to 1335008 bits (step 301).

Next, the control circuit 5 sets the target bit rate R
Is divided by the frame rate to obtain the target bit generation amount Tpic per picture, and the target bit rate R is divided by the frame rate and the number of each macroblock in one picture to obtain the target bit rate per macroblock. Obtain the generation amount Tmb (step 3
02).

Subsequently, if the target bit rate R is less than 3 Mbps (step 303, Ye
s), and sets the reference quantization width q_st to 6 (step 30).
4), if the target bit rate R (bps) is less than 6 Mbps (Step 305, Yes), the reference quantization width q_st
Is set to 5 (step 306), and the target bit rate R
If (bps) is 6 Mbps or more (step 30)
5, No), and sets the reference quantization width q_st to 4 (step 307).

After setting the reference quantization width q_st according to the target bit rate R in this way, the control circuit 5 sets the target bit rate R and the reference quantization width q_st to the following equations (14),
Substituting into (15), (16) and (17), each coefficient K
1, K2, K3, K4 are obtained, and these coefficients K1, K
2, K3, K4, target bit rate R and reference quantization width q_
The function f (R, Dmb) is set by substituting st into the following equations (18), (19), and (20) (step 308). K1 = R × (q_st−2) / (2 × 500000) (14) K2 = 3R / ((q_st + 3) × 500000) (15) K3 = R × (70−q_st) / (70 × 800000) (16) K4 = q_st × R / (R−K3 × 500000) −q_st × R / (R−K2 × 500000) (17) When Dmb <0 f (R, Dmb) = q_st × R / (R −K1 × Dmb) (18) When 0 ≦ Dmb <500000 f (R, Dmb) = q_st × R / (R−K2 × Dmb) (19) When Dmb> 500000 f (R, Dmb) = q_st × R / (R−K3 × Dmb) −K4 (20) where the quantization width q_sc derived from the function f (R, Dmb)
Rounds ale after the decimal point. Also, the quantization width q_sca
If le is 1 or less, set the quantization width q_scale to 1,
If the quantization width q_scale is 31 or more, the quantization width q_scale
Is set to 31.

Thus, the above equations (18), (19) and (2)
After setting the function f (R, Dmb) shown in (0), the control circuit 5 repeats the processing of each of steps 102 to 109, 110A, and 111.
By performing an operation based on (R, Dmb), the quantization width q_scale
To update.

As described above, by appropriately setting the reference quantization width q_st according to the target bit rate R, the target bit rate R
, It is possible to obtain a reproduced image having a stable image quality according to.
For example, when the target bit rate is lower than 3 Mbps, the average quantization width is large. Therefore, a function f () in which the quantization width changes at a small rate of change near this average quantization width. R, Dmb), the image quality is stabilized. Similarly, when the target bit rate is as high as less than 6 Mbps, the average quantization width is small. Therefore, a function f such that the quantization width changes at a small rate of change around this average quantization width. If (R, Dmb) is set, the image quality of the reproduced image is stabilized.

FIG. 12 is a flowchart showing an encoding process according to the third embodiment of the present invention. This encoding process is performed in the encoding processing device shown in FIGS. 1, 3, and 5.

The processing of the flowchart of FIG. 12 is performed by the steps 101 and 1
10 is replaced by steps 101B and 110B, and step 401 is further added.
02 to 109 and 111 are common to the processing of the flowchart of FIG.

Therefore, the processing of the flowchart of FIG. 12 differs from the processing of the flowchart of FIG. 6 in the initialization processing of each parameter in step 101B and the derivation processing of the quantization width q_scale in step 110B. The update processing of the coefficient K5 in 401 is different.

In step 101B, the control circuit 5 of the rate control unit 3 sets the target bit rate R (bps) to a predetermined value and sets the size of the VBV buffer to 1
835008 bits are set, and the bit generation error Dmb is set to 0.
And the occupation amount VBV_fullness is set to 935008 bits. Further, by dividing the target bit rate R by the frame rate, the target bit generation amount Tpic per picture is obtained, and the target bit rate R
Is divided by the frame rate and the number of macroblocks in one picture, thereby obtaining a target bit generation amount Tmb per macroblock. Further, the coefficient K5 is set to 0, the reference quantization width q_st is set to 3, and the immediately preceding quantization width last_q and the quantization width q_scale are both set to 3, which is the same as the reference quantization width q_st.

This step 101B determines the occupation amount VBV_f.
The step of FIG. 6 is that the ullness is set to 935008 bits, the coefficient K5 and the immediately preceding quantization width last_q are added to two new parameters, and the reference quantization width q_st is set to a lower value of 3. Differs from 101.

After that, the control circuit 5 determines in each step 102
~ 107, 401, 108, 109, 110B, 111
Is repeated, and each time, the actual bit generation amount Bmb per macroblock is obtained (step 106).
The bit generation error Dmb is updated based on the above equation (3) (step 107), the coefficient K5 and the immediately preceding quantization width last_q are updated (step 401), and the bit generation error Dmb does not exceed the threshold value D4. (Step 1
08, No) and an operation based on the function f (last_q, K5, Dmb) to update the quantization width q_scale (step 11).
0B), and quantizes each transform coefficient obtained by the DCT transform according to this quantization width q_scale (step 111),
The quantized data is subjected to variable-length encoding to form compression-encoded image data.

Next, the processing of step 401 will be described with reference to the flowchart of FIG. First, the control circuit 5 updates the immediately preceding quantization width last_q to the quantization width q_scale used when encoding the immediately preceding macroblock (step 411), and determines whether or not the bit generation error Dmb> 5000000 Is determined (step 412). If the bit generation error Dmb> 500000 (step 412, Ye
s), the control circuit 5 determines that the immediately preceding quantization width last_q is 7 or more and 1
After confirming that the coefficient is within the range of 0 or less (step 413, Yes), the coefficient K5 is set to the immediately preceding quantization width last_q.
(7 or more and 10 or less) minus 6 (1 or more and 4 or less)
(Step 414). However, if the immediately preceding quantization width last_q does not fall within the range of 7 or more and 10 or less (No in Step 413), the coefficient K5 is not updated in Step 414.

Also, the bit generation error Dmb> 500000
If not (step 412, No), the control circuit 5
≧ Bit generation error Dmb> −100000 and coefficient K5 ≧
It is determined whether it is 4 (step 415), and if this condition is satisfied (step 415, Yes), the coefficient K5 is updated to 3 (step 416). If this condition is not satisfied (step 415, No), the procedure moves to the next step 417.

In step 417, the control circuit 5
Is −100000 ≧ bit generation error Dmb> −2000
It is determined whether 00 and the coefficient K5 ≧ 3. If this condition is satisfied (step 417, Yes), the coefficient K5 is updated to 2 (step 418). If this condition is not satisfied (step 417, No), the procedure moves to the next step 419.

In step 419, the control circuit 5
Is -200000 ≧ bit generation error Dmb> −3000
It is determined whether 00 and the coefficient K5 ≧ 2, and if this condition is satisfied (Step 419, Yes), the coefficient K5 is updated to 1 (Step 420). If this condition is not satisfied (step 419, No), the procedure moves to the next step 421.

In this step 421, the control circuit 5
Determines whether the bit generation error Dmb <−300000, and if this condition is satisfied (step 421, Ye
s), the coefficient K5 is updated to 0 (step 422), and if this condition is not satisfied (step 421, No), the coefficient K5
Do not update

That is, in the processing of the flowchart in FIG. 13, the immediately preceding quantization width last_q is updated to the quantization width q_scale used when encoding the immediately preceding macroblock. Steps 412, 415, 41
If the bit generation error Dmb is not included in any of the numerical ranges shown in FIGS.
If 0 ≧ bit generation error Dmb> 0, the coefficient K5 is not updated. Further, when the bit generation error Dmb> 500000, the coefficient K5 is set to the immediately preceding quantization width last_q (7 or more and 10 or less) on the assumption that the immediately preceding quantization width last_q is in the range of 7 or more and 10 or less. Is updated to the difference of 1 minus 4 (1 or more and 4 or less). Thereafter, when the bit generation error Dmb decreases to 0 or less, the coefficient K5 is returned to 0 stepwise according to the degree of the reduction.

After updating the immediately preceding quantization width last_q and setting the coefficient K5 in this manner, in step 110B of the flowchart of FIG. 12, the function f (last_q,
K5, Dmb) to update the quantization width q_scale.

FIG. 14 shows characteristics of a function f (last_q, K5, Dmb) for deriving the quantization width q_scale.

A function f (last_q, K5,
Dmb) are represented by the following equations (21) to (28).

When Dmb <0, f (last_q, K5, Dmb) = 1800000 / (600000−Dmb) + K5 (21) When 0 ≦ Dmb <500000, f (last_q, K5, Dmb) = 300000 / (1000000−) Dmb) + K5 (22) Dmb ≧ 500,000 and 3000000 / (1000000−Dmb) <7 and K5 = 1 f (last_q, K5, Dmb) = 7 (23) Dmb ≧ 500,000 and 3,000,000 / (1000000−Dmb) ) <8 and K5 = 2 f (last_q, K5, Dmb) = 8 (24) Dmb ≧ 500000 and 300000 / (1000000−Dmb) <9 and K5 = 3 f (last_q, K5, Dmb) = 9 ... (25) Dmb ≧ 500000 and 300000 / (1000000−Dmb) <10 and K5 = 4 f (last_q, K5, Dmb) = 10 (26) Except for the above conditions, 900000 ≧ Dmb ≧ 500000 Case f (last_q, K5, Dmb) = 300000 / (1000000−Dmb) (27) Except for the above conditions, when 900000> Dmb, f (last_q, Dmb) K5, Dmb) = 31 (28) where the decimal part of the quantization width q_scale derived from f (last_q, K5, Dmb) is rounded off.

As described above, after the coefficient K5 is initialized to 0 (step 101B), the coefficient K5 is not updated as long as 500000 ≧ bit generation error Dmb> 0 (step 401). The quantization width q_scale is
It is derived from the above equation (22), changes along the characteristic F1 in the graph of FIG. 14, and matches the reference quantization width q_st = 3 when the bit generation error Dmb = 0.

Thereafter, for example, the bit generation error Dmb> 5
When the value becomes 00000 (step 412, Yes) and the immediately preceding quantization width last_q = 10 (step 413), the coefficient K5 is updated to 4 (step 414).

Next, when the quantization width q_scale is adjusted and the bit generation error Dmb decreases, the quantization width q_scal
e is derived from the above equation (26) and changes along the characteristic F9 in the graph of FIG. Furthermore, 500000 ≧
When the bit generation error Dmb> 0, the quantization width q_scale
Is derived from the above equation (22), but the coefficient K5 is 4
To the characteristic F5 in the graph of FIG.
Varies along.

Next, the bit generation error Dmb decreases to 0
≧ Bit generation error Dmb> −100000 and coefficient K5 ≧
If it is determined to be 4 (Step 415, Yes), the coefficient K5 is updated to 3 (Step 416), and the quantization width q_scal
e is derived from the above equation (22) and changes along the characteristic F4 in the graph of FIG.

Thereafter, 500000 ≧ bit generation error D
If mb> 0, the quantization width q_scale changes along the characteristic F4 in the graph of FIG. Further, the bit generation error Dmb further decreases, and the bit generation error Dmb <−30.
0000 (step 421, Ye
s) Since the coefficient K5 is updated to 0 (step 422),
The quantization width q_scale is derived from the above equation (22),
It changes along the characteristic F1 in the graph of FIG.

Similarly, the bit generation error Dmb> 500
000, the previous quantization width last_q is 7
Assuming that the coefficient falls within the range of 10 or more, the coefficient K
5 is updated to be larger, and after this, 500000 ≧
When the bit generation error Dmb> 0, one of the characteristics F2 to F5 is selected according to the magnitude of the coefficient K5, and as the bit generation error Dmb becomes smaller, the coefficient K5 is updated stepwise to become smaller. Return to the characteristic F1.

That is, in the processing of the flowchart of FIG. 12, a characteristic that draws a hysteresis loop is applied as the characteristic of the quantization width q_scale. For this reason, the degree of difficulty in coding the picture increases, the amount of bits generated for coding increases, the bit generation error Dmb increases, and when the quantization width q_scale is sufficiently large, the hysteresis loop According to the upper characteristic curve of
Since the quantization width q_scale changes, this quantization width q_scale
Can be maintained large, and the image quality of the reproduced image can be stabilized.

Further, after the difficulty of encoding becomes extremely high and k5 becomes large, a scene change or the like causes
When the encoding difficulty becomes relatively low, the bit generation error Dmb rapidly decreases, and in a region where the bit generation error Dmb <0, the quantization width q_scale rapidly decreases, but the direction of increase thereafter. In this case, since the rate of change of the quantization width q_scale becomes small, the image quality is improved, and then this image quality is stably maintained.

As described above, when the degree of difficulty of encoding is low, the quantization width is set to be small to keep the image quality of the reproduced image stable, and when the degree of difficulty is high, the quantization width is increased. By maintaining this, the image quality of the reproduced image is stabilized, so that it is not necessary to change the function for deriving the quantization width for each target bit rate, and the image quality is stabilized according to the difficulty of encoding. Such encoding can be performed.

In the third embodiment, the characteristic curve as shown in the above equation is derived as a function for deriving the quantization width. However, as a characteristic for deriving the quantization width, a hysteresis loop may be drawn. For example, the characteristic curve may be approximated to a straight line. Here, the quantization width and the bit generation error are used as conditions for the hysteresis characteristic, but one can be derived from the other based on the relationship between the quantization width and the bit generation error. May be used.

[0118]

As is apparent from the above description, according to the present invention, a bit generation error which is a difference between an actual bit generation amount and the target bit generation amount is obtained, and a quantum corresponding to the bit generation error is determined. In the predetermined range of the bit generation error, a characteristic in which the quantization width changes small with respect to the change in the bit generation error is set, and the quantization for the bit generation error is set based on this characteristic. Seeking the extent of change. For this reason, in a state where the bit generation error changes within a predetermined range of the bit generation error, the quantization width does not largely change, and any of the pictures I, P, and B in the GOP is not changed. , An appropriate target bit generation amount is allocated, and the image quality of each reproduced image corresponding to these pictures is stabilized. Also, in one picture, as long as the bit generation error changes within a predetermined range of the bit generation error, an appropriate quantization width is set, and the degree of difficulty in coding each macroblock is set. Thus, an appropriate bit generation amount can be allocated to each macroblock.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a first embodiment of an image compression encoding apparatus according to the present invention.

FIG. 2 is a diagram used for explaining an outline of a video signal encoding process;

FIG. 3 is a block diagram showing a basic encoding processing unit in the apparatus of FIG. 1;

FIG. 4 shows VB when encoding a video signal at a fixed rate.
Diagram showing data occupancy in V buffer

FIG. 5 is a block diagram showing a rate control unit in the device of FIG. 1;

FIG. 6 is a flowchart showing an encoding process in the apparatus of FIG. 1;

FIG. 7 is a flowchart showing a stuffing process of step 104 in the encoding process of FIG. 6;

8 is a graph showing characteristics of a function f (R, Dmb) for deriving a quantization width q_scale in the apparatus of FIG.

FIG. 9 is a graph showing characteristics of another function f (R, Dmb) for deriving a quantization width q_scale in the apparatus of FIG. 1;

FIG. 10 is a flowchart illustrating an encoding process according to the second embodiment of the present invention.

11 is a step 101 in the encoding processing of FIG. 10;
Flow chart showing initialization processing of A

FIG. 12 is a flowchart illustrating an encoding process according to a third embodiment of the present invention.

FIG. 13 is a step 401 in the encoding process of FIG. 12;
Flowchart showing the processing of

FIG. 14 is a graph showing characteristics of a function function f (last_q, K5, Dmb) for deriving a quantization width q_scale in the third embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Basic encoding processing part 2 Encoder sending buffer 3 Rate control part 4 MB bit counter 5 Control circuit 11 Image rearrangement part 12 Scan conversion processing part 13 Difference part 14 DCT transformation processing part 15 Weighting quantization processing part 16 Variable length code Processing section 17 inverse quantization section 18 inverse DCT processing section 19 prediction processing section with motion compensation 20 mode determination section 21 motion detection processing section 22 addition processing section

Claims (14)

[Claims] An image compression encoding apparatus for determining a quantization width when encoding a video signal, and encoding the video signal based on the quantization width, wherein the video compression encoding device is used for encoding the video signal. In addition to obtaining the actual bit generation amount which has been reduced, a target bit generation amount corresponding to a target bit rate is obtained, and a bit generation error which is a difference between the actual bit generation amount and the target bit generation amount is obtained. A quantization width corresponding to the generation error is obtained, and in a predetermined range of the bit generation error, a characteristic in which the quantization width changes small with respect to a change in the bit generation error is set, and based on this characteristic, An image compression encoding apparatus comprising control means for calculating a quantization width for a bit generation error by using the control means. 2. A reference quantization width is predetermined, and the control means obtains a ratio of a value obtained by subtracting a value corresponding to a bit generation error from a target bit rate to the target bit rate, and calculates the ratio and the target bit rate. 2. The image compression encoding apparatus according to claim 1, wherein a reference quantization width is integrated, and a quantization width for encoding a video signal is obtained based on the product. 3. The control means obtains a reference quantization width based on a target bit rate, and in a predetermined range of a bit generation error, the quantization width is changed with respect to a change in the bit generation error. 2. The image compression encoding apparatus according to claim 1, wherein a characteristic that changes slightly near the width is set, and a quantization width for a bit generation error is determined based on the characteristic. 4. The image compression coding according to claim 1, wherein the characteristic of the change of the quantization width with respect to the change of the bit generation error draws a hysteresis loop in a predetermined range of the bit generation error and in the vicinity of the predetermined range. apparatus. 5. A virtual buffer for sequentially inputting image data obtained by encoding video signals and sequentially outputting the image data, the occupancy of the virtual buffer corresponds to a bit generation error. When the occupation amount of the virtual buffer approaches 0, the quantization width sharply approaches the maximum value, and the quantization width is changed so as to be linked to the characteristic of the quantization width that changes small with a change in the bit generation error. The image compression encoding apparatus according to claim 1, wherein the width changes continuously. 6. A plurality of ranges of the bit generation error are predetermined, and a characteristic indicating a change in the quantization width with respect to the change of the bit generation error is set for each of the ranges. 2. The image compression encoding apparatus according to claim 1, wherein a range in which the bit generation error is included is selected from, and a quantization width for the bit generation error is obtained based on characteristics of the range. 7. A video signal is encoded in macroblock units based on a quantization width in accordance with the MPEG standard, and the control means calculates a quantization width corresponding to a bit generation error for each macroblock. The image compression encoding apparatus according to claim 1. 8. An image compression encoding method for determining a quantization width when encoding a video signal and encoding the video signal based on the quantization width, the method comprising: In addition to obtaining the actual bit generation amount which has been reduced, a target bit generation amount corresponding to a target bit rate is obtained, and a bit generation error which is a difference between the actual bit generation amount and the target bit generation amount is obtained. A quantization width corresponding to the generation error is obtained, and in a predetermined range of the bit generation error, a characteristic in which the quantization width changes small with respect to a change in the bit generation error is set, and based on this characteristic, Image compression encoding method for obtaining a quantization width for a bit generation error using 9. A reference quantization width is predetermined, and a ratio between a value obtained by subtracting a value corresponding to a bit generation error from a target bit rate and the target bit rate is determined. 9. The image compression encoding method according to claim 8, wherein a quantization width for encoding a video signal is obtained based on the product. 10. A reference quantization width is determined based on a target bit rate, and in a predetermined range of a bit generation error, the quantization width is small near the reference quantization width with respect to a change in the bit generation error. 9. The image compression encoding method according to claim 8, wherein a changing characteristic is set, and a quantization width for a bit generation error is obtained based on the characteristic. 11. The image compression encoding method according to claim 8, wherein within a predetermined range of the bit generation error, the characteristic of the change of the quantization width with respect to the change of the bit generation error draws a hysteresis loop. 12. When assuming a virtual buffer for sequentially inputting each image data obtained by encoding a video signal and sequentially outputting these image data, the occupation amount of the virtual buffer corresponds to a bit generation error. When the occupation amount of the virtual buffer approaches 0, the quantization width sharply approaches the maximum value, and the quantization width is changed so as to be linked to the characteristic of the quantization width that changes small with a change in the bit generation error. 9. The image compression encoding method according to claim 8, wherein the width changes continuously. 13. A plurality of ranges of the bit generation error are predetermined, and a characteristic indicating a change in the quantization width with respect to the change of the bit generation error is set for each of these ranges. 9. The image compression encoding method according to claim 8, wherein a range in which is included is selected, and a quantization width for a bit generation error is obtained based on characteristics of the range. 14. A video signal is encoded in macroblock units based on a quantization width in accordance with the MPEG standard, and a quantization width corresponding to a bit generation error is obtained for each macroblock. Image compression encoding method as described in the above.