KR100531028B1 - Process for coding digital data of a video image sequence - Google Patents

Abstract

The present invention relates to a method of coding a video sequence, which is digital data, for transmission at a variable bit rate.

The coding method is characterized in that the bit rate adjustment for the image at the instant n is performed according to the correlation regarding the transmission channel bit rate prediction at the instant n + τ, where τ is It is a constant time between the instant stored in the coder's buffer and the instant outputted from the decoder buffer.

Description

PROCESS FOR CODING DIGITAL DATA OF A VIDEO IMAGE SEQUENCE

The present invention relates to a method of coding audio digital data, video digital data, or auxiliary digital data.

More particularly, the present invention relates to systems for transmitting and storing at variable bit rates using compression schemes for audio and video digital data. The present invention is compatible with the MPEG 2 standard.

The role of the video compression system is to achieve the best image quality at the lowest bit rate possible. First of all, the coding quality performed using this approach depends on the choice of motion estimator and coder regulating algorithm. The choice of motion estimator and coder adjustment algorithm depends directly on the optimality criteria adopted.

In terms of motion estimation, the search performance for the reference image is an important characteristic. The main goal of the regulation is to ensure that the coded image maintains a stable quality. The "padding" factor, which is adopted to adjust the coding to match the output bit rate (usually inserts "0" in the data stream to make it possible to maintain the set bit rate), is optimized for To be minimized.

The cost of coding each image is variable. The role of the regulating loop is to correct for variations in the data flow due to variations in image complexity (defined below) and to transmit the data flow corresponding to the data rate setting as an output. This adjustment generally affects the quantization interval, which is determined by the fill level of the buffer memory. It is possible to alleviate the variation in the bit rate caused by the response time of the adjustment loop of the buffer memory at the output side of the coder, and the capacity of the buffer is related to the variation in the bit rate.

The data flow at the coder output of the video source may be variable. For example, in the case of transmitting several video sources over a single channel, these channels are available "global" dynamically as a function of the change in the complexity of the images to be coded from each video source. Allocating bitrates can actually be beneficial. The method makes it possible to improve the overall quality of the transmitted images as compared to existing methods such as assigning bit rates to video sources as a function of the transmitted program type.

1 illustrates a set of transmissions using this method. The transmission includes transmitting at a constant global bit rate from several coded video sources over a single transport channel. This bit rate corresponds to the capacity of the channel, for example the capacity of the transponder channel. The coding circuits on the video source side 13i are linked to the input side of the multiplexer 11, and the output of the multiplexer supplies a data stream transmitted over the channel. The coding circuits on the video source side are also linked to the interface circuit 14, which transmits data pertaining to the complexity of the video sources to the bit rate allocator 12, and also by the bit rate allocator 12 Receive the bit rates assigned to each source. Therefore, the bit rate allocator, which serves to allocate a bit rate to each source 13i, controls the multiplexing circuit, that is, the multiplexer 11. Information is exchanged using a fast bus connected to this bit rate allocator 12 and interface circuit 14. Supervisory circuit 15 is linked to the bus to manage the transmission set.

Multiplexing is performed as a function of the bit rate assigned to each video source transmitted over the same channel, for example by the bit rate allocator at the GOP rate, said GOP being a group of images defined below. The bit rate is dependent on the complexity coefficient of the video source, which takes into account the amount of information generated during a given quantization interval. The complexity coefficient is measured for each GOP of each video source. If the quantization interval is constant, then the complexity coefficient corresponds to the number of bits used to code the GOP, for example. Alternatively, if this quantization step is variable, then the complexity coefficient is multiplied by the number of bits multiplied by the quantization interval. Corresponding to the corresponding standardized bit counting. Since the data flow is constant during one GOP period, the output buffer at the corresponding source must mitigate the bit rate spikes that appear during one GOP period.

The use of variable bit rate encoders, associated with statistical multiplexing depending on the bit rate of each source, is known as described above and is a means of optimizing the capacity of the transponder channel. The disadvantages of these adjustment methods are also known. These disadvantages are mainly the size of the buffer memory in the decoder, which needs to be large in order to be able to absorb bit rate variations of the corresponding video sources without running out of memory or saturation of the memory. The decoder's buffer must neither be overflowed (loss of data) nor exhausted. That is why the cost of the decoder has increased. In other respects, the amount of memory required is generally no longer compatible with the MPEG 2 standard.

Whether bit rate instantaneous fluctuations occur at the coder side or at the decoder side or alternatively should be limited as a function of these buffer memory capacities, the instantaneous variation of the bit rate, i.e. image level, is much larger. Doses are generally needed.

1 shows a transmission set known in the prior art;

2 shows an apparatus for coding and decoding according to the MPEG2 standard.

3 schematically illustrates the fullness of the coder and decoder buffer memory as a function of bit rate;

4 is a diagram showing an example of change in position and size of a virtual buffer as a time function;

5 shows the position and size of a virtual buffer in the coder's physical buffer as a function of bit rate.

6 shows the size of a virtual buffer in the coder's physical buffer as a function of bit rate.

Fig. 7 shows the change of the parameter α as a function of the bit rate.

8 illustrates an apparatus for coding in accordance with the present invention.

9 illustrates an apparatus for regulation in accordance with the present invention.

10 is a flow chart showing an adjustment method according to the present invention.

The object of the present invention is to eliminate the above mentioned drawbacks.

SUMMARY OF THE INVENTION An aspect of the present invention is a method of coding digital data of a video image sequence, which realizes compression of the data image and adjustment of the bit rate at the coder output using an adjustment loop that operates on the quantization interval of the data to be coded, In order to enable the coder to transmit data through the channel at a variable bit rate to the buffer side, the coder stores the output data by the coder in the buffer, and outputs the image from the buffer at the instantaneous time and the image stored in the buffer of the coder. Realizing the data tagging to ensure that there is a constant time τ between instants, and the bit rate adjustment for the image at the instant Implemented as a function of bit rate prediction, and the regulation is performed by the coder buffer. Runs within the limits of a virtual buffer defined by a top address and a bottom address, before the time τ of a change in the transmission bit rate at the output of the coder buffer occurs. And depending on the value corresponding to the bit rate variation multiplied by time [tau], the upper address of the virtual buffer, i.e. the offset, varies according to the upper address of the coder buffer.

Another aspect of the present invention is a method of transmitting data from several video sources, which realizes data coding for each source in order to transmit data at a variable bit rate over a given transmission channel, the channel available bit rate and a set of And a large bit rate allocation to each source before a time τ of the actual bit rate of the source appears through the channel depending on the complexity measurement of the preceding image from the video sources.

The features and advantages of the present invention are better illustrated by the following description given by reference to the accompanying drawings and examples.

An advantage of the present invention relates to the reduced size of the buffer memory in the decoder. This memory is therefore intended to be compatible with the MPEG 2 standard while allowing instantaneous bit rate changes within a very large range. The method ensures a decoder that adapts to the size limitations of the video buffer, but has the ability to respond to changes in each image, bit rate. This performance is most notable in that it does not introduce any limit on the maximum step in bit rate change, namely 1.5-15 Mbit / s bracket of MP @ ML MPEG2 format (MPEG2 main profile main level).

Due to the advantages of the present invention, the available bit rate of the channel is optimized, i.e. assigned in such a way that the value and time are optimized as a function of the respective video source information complexity.

The principle of video compression, commonly referred to as MPEG2, is the subject of international recommendations (ISO / IEC 13818 H.262). Since the functional structure of the MPEG2 compressor is known as described in the MPEG2 standard, only a brief description will be given below.

The coder's video input receives digitally coded images in 4.2.2 format. In the case of spatial coding (intra image or "intra" coding), the images, or more precisely the blocks that make up the image, are directly subjected to a discrete cosine transform (DCT) followed by a quantization process (image I). The quantized values are then variable length coded (VLC) before being stored in memory or buffer. The latter provides "smoothing" of the bit rate at the output of the coder, and the adjustment loop also makes it possible to link the quantization interval to the bit rate setting. The inverse quantization process, followed by an inverse DCT transform, is also performed in such a way as to store a reconstructed image in memory, which plays a support role in motion estimation.

Inter-image, or "inter" coding, can take advantage of temporal redundancy by coding shifting between images for blocks. The two forms of temporal coding are:

Forward predictive coding (P image) that solely relies on only images of a similar or self-like form among preceding images during motion estimation;

Bidirectional coding (B image) depending on the images (I and / or P images) directly contacting the image to be coded.

The cosine transform only applies to the differences between the images in the case of time coding. The quantized error accompanying the vector thus allows the reconstruction of the data at the decoder, which characterizes the movement of the blocks making up the image.

The set in which the images (P image, B image) preceding the next image (I image) is made up of the subsequent image (I image) is called GOP (Group Of Pictures).

An explanation of the principle of bit rate adjustment, that is, the gist of the present invention, refers to the VBV model as described in the MPEG2 standard.

The VBV model, or Video Buffering Verifier, can be defined as a virtual decoder connected to the output of the coder, which makes it possible to model constraints on variations in bit rate. This is under the assumption that the decoder instantaneously extracts the images to be decoded from its buffer. The VBV model is present in the algorithms to adjust the coders and is also essential for the coder to know exactly the instantaneous value at which these images are output from the decoder's buffer. It is the coder itself that is responsible for determining these decoding instants using the method described in FIG. 2 according to the MPEG2 standard.

The coder 21 consists of the following components:

A coding module 211 for receiving digital video data at its input, that is, at the input of the coder,

An output buffer 212 which receives at its input the coded data starting from the coding module,

The resulting bit stream or data flux is transmitted over the transmission channel 22, cable, wireless link or satellite communication link, and also at its output at the output of the buffer 212 to insert a clock reference to the data starting from the buffer. An insertion circuit 213 to which an input of is linked, and

The coder receives a synchronization signal corresponding to the instant of arrival of the image from the coding module 211, sends a Decoding Time Stamp (DTS) "decoding instant" according to the MPEG 2 standard to the module side, and also to the insertion circuit side Internal clock circuitry 214 for supplying a Program Clock Reference (PCR) clock value.

Data transmitted through the channel reaches the input side of the decoder 23. The decoder includes the following components:

A demultiplexing circuit 231 which performs PCR clock reference extraction and also receives data transmitted through its channel at its input, ie at the input of the decoder,

A buffer 232 of the decoder for receiving data originating from the demultiplexing circuit,

A decoding module 233 whose input is linked to the output of the buffer, which performs a decoding operation opposite to the coding operation and supplies decoded video data at its output, which is also the output of the decoder 23, and

An internal clock circuit 234 linked to a buffer for reading the DTS decoding instant and transmitting a read signal thereof and also linked to the demultiplexing circuit for receiving a PCR (Program Clock Reference) reference.

Thus, approximately every 100 ms, the coder's internal clock value (PCR) is transmitted on the MPEG2 bit stream at the level of the transport layer (as defined in the standard). The decoder extracts this PCR value to latch its time reference using a phase locked loop, or PLL, located in the internal clock circuit 234. The coder and decoder thus have a common reference clock.

For each image provided at the input of the coder, the PCR clock is sampled and compensated by the transit time. The obtained information, i.e. the decoding instant tag (DTS = decoding time stamp), is inserted into the MPEG2 bit stream associated with the image and also corresponds to the programmed decoding instant.

The decoder is responsible for receiving the images in its calibration buffer, which stores the DTS instants for each of the images, at which time the coder programmed the decoding. Therefore, whenever the PCR clock data reconstructed at the decoder reaches the DTS value of the image on the output side of the buffer, the clock circuit initializes the decoding, and the content of the bit stream corresponding to the image is read using a read signal. Extracted from the decoder's buffer.

While the buffer at the decoder is full (VBV_fullness = VBV_size), when the decoder first assumes that the encoded image should be extracted from the buffer, if VBV_ size is the size of the decoder buffer, then VBV_fullness If is the fill level of the decoder buffer at a given moment, the decoder must wait for the same time before enabling its output side:

Where τ is the transition time in the buffers (end-to-end delay)

The rate is the output bit rate of the coder (transmission bit rate).

This is actually the time the image is in the coder buffer.

A simplified diagram for the complete coding / decoding chain is shown in FIG.

The following notation has been adopted.

D1 (t) = bit rate (input of the coder buffer) when leaving the coding circuit at the instant t (t),

D2 (t) = bit rate (output of the decoder buffer) when entering the decoding circuit at the instant t,

B1 (t) = coder's buffer occupancy at the instant (t),

B2 (t) = decoder's buffer occupancy at instant t

D _C (t) = channel bit rate at instant t.

The coding circuit 31 which receives the video data at its input is linked to the buffer memory 32 side which is symbolized by the reception container filled with the data amount B1 (t) which is the occupancy of the buffer. This buffer is linked to the input side of the decoder buffer 34 by a physical link, a wireless communication link, or a satellite communication link, which is referred to as transport channel 33. Each buffer 34 has a fullness {B2 (t)}. The output of the buffer 34 is suitably linked to the input of the decoding circuit 35.

τ1 is the image residence time in the buffer 32, τ2 is the image residence time in the buffer 34, τ is the instant at which the image is input into the coder's buffer and the buffer of the decoder Is the elapsed time between the instantaneous time from which the image is output.

In order to ensure that the global processing time (compression + storage + transmission + storage + decompression) is constant, each image must pass through the buffering and storage stages of the coder and decoder in a certain amount of time (τ). The decoding time itself is constant, and the bit rates at the input side of the coder and the output side of the decoder are the same: τ = τ 1 + τ 2 = constant value.

Referring to FIG. 2, in order to guarantee a constant time τ, the decoding instants, which are inserted into the bit stream and denoted by DTS, may be configured to have a sampling value for the PCR clock corresponding to the time when the image arrives at the input of the coder. It is calculated by having a positive offset equal to the transition time τ (transition time referred to as end-to-end delay) of the decoder buffer.

Since τ is the transmission time for the complete image, the bit rate of the input of the coder buffer at instant t is equal to the bit rate of the output of the decoder buffer at instant t (τ):

D1 (t) = D2 (t + τ)

The initial conditions at the instant t0 are as follows:

B1 (t0) = B2 (t0) = 0

At the instant (t ₀ + τ), the decoder's buffer state is:

At any instant, the coder's buffer state is:

Similarly, the decoder buffer state at the instant t + τ can be written as:

Thus, the following equation can be obtained:

The above relation defines the behavior of the decoder buffer based on the values measured in the coder buffer.

When the bit rate on the channel is constant, the buffer states are complementary:

B ₂ (t + τ) = B ₂ (t ₀ + τ) -B ₁ (t), where B ₂ (t ₀ + τ) is the initial state of the decoder buffer

This result represents an intuitive principle for communicating vessels between these two buffers, with a delay time that matches the pass time of the data item passing through the encoder buffer and decoder buffers.

In MPEG2's main profile main level (MP @ ML) video coding, the constant value obtained by summing two buffer amounts is limited to 1.835 Mbit.

When the bit rate in the channel is variable, the state of the decoder buffer can be clearly seen under conditions with the following amounts:

This amount equal to B1 (t) + B2 (t + τ) is located in the coder buffer and specifies the offset of the virtual buffer as described below.

Fig. 4 shows the accumulated occupancy of the buffer—the amount of information accumulated in the buffer of the coder and the buffer of the decoder—as a function of time, making it possible to infer the change in fill of each buffer.

The horizontal axis represents time and the vertical axis represents the amount of information accumulated in the buffers, i.e., the accumulated occupancy rate.

The curve 41 may be referred to as a coder buffer, either to a buffer write address or to a buffer write pointer, where each vertical stroke is a representation of the information stored by the buffer, which is necessary for intercoding or intracoding the image. Amount (the length of the stroke is longer in the case of intra coding).

Curve 42 is for the decoder buffer, decoder buffer read pointer, and represents the amount of accumulated information starting from the buffer. By the relationship described above, the second graph is moved by the value τ along the longitudinal axis of the first curve.

Curve 43 shows the amount of accumulated information starting from the coder buffer and the amount of information input to the decoder buffer. These graphs overlap and represent the amount of information transmitted over the channel since the transit time in the channel was assumed to be zero to simplify the proof. In fact, graph 43 represents the read pointer of the coder buffer, which is overlaid onto the write pointer of the decoder buffer. The slope of the curve is the channel bit rate Dc (t), that is, the bit rate at the output of the coder buffer and the bit rate at the input of the decoder buffer.

Curve 44 represents the actual size of the coder buffer, and curve 45 represents the size of the decoder buffer. This is a shift of the curve 43 along the longitudinal axis. In the former case, the coder buffer is moved by a positive vector modulated by the modulus, and in the latter case, the negative vector is obtained by using the size of the decoder buffer as the coefficient. As much as I moved. Finally, curves 46 and 47 contribute to modeling the virtual buffer in the coder buffer by defining the location and size of the virtual buffer.

The virtual buffer represents a "useful" adjustment area in the coder buffer, which in the VBV model constitutes an image.

Positioning in the coder's physical buffer is calculated to ensure that the transition time for each image is constant at all bit rates.

This is defined by the parameter VBV_offset which is the maximum address in the adjustment area at a predetermined instant t. This corresponds to the deviation between curve 43 and curve 46 and can be read on the longitudinal axis.

Knowing the parameter VBV_offset makes it possible to determine whether at all instants the decoder has complied with the buffer occupancy constraints in the case of the VBV model. It is necessary to know this in order to function the tuning algorithm appropriately.

The positioning of the virtual buffer in the coder's physical buffer can be obtained as shown below.

The relation is as follows:

The relation can be represented by "discretizing" at the image level:

In the discrete formula:

n represents the image at the instant t

Dc (k) represents bits per image at the output side of the coder while coding image k.

τ is represented by the number of images during the duration τ.

Thus, the expression for the previous image can be obtained as follows:

From this the variation in the buffer occupancy factor is obtained:

Note the following here:

ΔB ₁ (n) = ΔD ₁ (n) -D _c (n) -ε _{Dc (n)}

We get a formula to update the occupancy factor of the virtual buffer:

ΔB ₂ (n + τ) = D _c (n + τ) -ΔD ₁ (n) + ε _{Dc (n)}

ΔD ₁ (n) in the above formula corresponds to the image coding cost at instant n, and D _c (n + τ) is the expected bit rate at the coder output side at the instant n. The third term in the equation is the error between the actual measured value Dc (n) measured at the instant n and the expected bit rate at the coder output.

Thus, the cost of image coding at instant n depends on the expected bit rate at the coder output side at instant n + τ. According to the above basic principle, the buffer occupancy information of the VBV model can be obtained by calculating the following at any instant:

VBV_fullness (n) = VBV_fullness (n-1) + Dc (n + τ)-Last_picture_cost + εDc (n)

According to the above conditions, the setting value is received just before coding the image n and started at an external device (bit rate allocator).

The bit rate allocator receives the complexity of the preceding image, for example during the GOP duration, to predict the channel bit rate after the instant τ, the complexity taking into account the complexity of the video sources that are required to be transmitted over the same channel. .

If Bpp (n) denotes the expected channel bitrate setting at the coder output during the image period n + τ, then it is possible to limit the update of the VBV model to the image level by the following equation:

VBV_fullness (n) = VBV_fullness (n-1) + Bpp (n)-Last_picture_cost + εBpp (n-τ)

here:

εBpp (n−τ) = Bpp (n−τ) −Dc (n), which is a setting received before coding the image (n−τ) and at the output side of the coder while coding the image (n). Indicates an error between the actual bit rates.

This relationship requires us to assume the main operating conditions of a variable bit rate coder:

The bitrate information signal received by the coordination algorithm must precede its actual application at the coder output by an amount of time equal to the transition time from the coder buffer to the decoder buffer.

The modeling of the decoder buffer state involves deductively recognizing the channel bit rate over the horizon [t, t + τ] at the moment of coding the image provided at the instant t with the coder.

Thus, by controlling the location of the virtual buffer, the coder can predict if any possible overstepping has occurred that may occur at the decoder. Referring to Fig. 4, the reproduction of the decoder buffer by the virtual buffer includes a prediction mechanism while the bit rate is changing.

For high bit rates, a useful area for adjusting the coder is located at the "top" part of the physical buffer indicated by curves 43 and 44, and at low bit rates it switches to the bottom of the buffer. .

This switching predicts that the bit rate of the channel changes by the duration [tau].

The passing time of the two buffers is constant. In order to operate at the minimum allowable bit rate, it is chosen by considering the pass time of the buffer. Here, τ is selected so that it is larger than VBV_size / bit rate_min.

If compliance with short delay constraints (possible constraints on the specification schedule) for high bit rates is required, the value chosen in this way can be disadvantageous. However, it is possible to limit the useful adjustment area to a low bit rate in order to reduce the time τ.

Finally note that the coder buffer size is larger than the decoder buffer size value by a ratio equal to the ratio of the maximum bit rate to the minimum bit rate. For certain types of images, the ratio will not exceed 4, thus making it possible to predict the coder buffer size of 8 Mbits (for 1.835 Mbits, which is a useful maximum size of the main profile main level).

5 shows a virtual buffer in the actual buffer of the encoder. This figure shows how the virtual buffer should be located and what its size should be determined as a function of the current bit rate.

The bit rate is reduced along the horizontal axis, and the unit is drawn using Mbits / s, and the "position" of the virtual buffer is Mbits, and a line is drawn in the ordinate.

In the maximum bit rate and saturation region, the size of the virtual buffer is equal to the size of the decoder buffer, which is 1.835 Mbits in this example.

Evbs = VBV_size

The virtual buffer is located on top of the encoder buffer.

In the linear domain, that is, between the 9 Mbits / s and 1.5 Mbits / s minimum bit rates in the example here, the buffer size decreases linearly so that:

Evbs = KR

α (n) = constant

Since the image level is still discontinuous, at the moment of coding image n, the size of the encoder's physical buffer (physical buffer size) is called Epbs (n), and the size of the encoder's virtual buffer (virtual buffer size) is It is called Evbs (n).

The virtual buffer is sized to be filled with a fully intra-coded image when the transmit bit rate is maximum.

The size of the virtual buffer of the encoder is at most the size of the decoder buffer, ie the so-called VBV_size; When the transmission bit rate is low, its magnitude can be reduced to the value {KR (n)}, where K is a time constant between 0.1 and 0.2, and the transmission bit rate R for the GOP structure and image n (n)}.

Evbs [n] = min (VBV_size, KR [n]) (a)

For simplicity, the size of the virtual buffer is consequently determined as a linear function of the bit rate until the size of the decoder buffer is reached at a given bit rate; The size of the buffer then remains constant at this saturation level.

Here is another example:

Epbs = 7.2 Mbits

K = 0.2 s

Rmax = 15 Mbit / s

Rmin = 1.5 Mbits / s

τ = 480 ms

Evbs are between 1.835 Mbits and at least 300 kbits.

The offset of the virtual buffer is 7.2 Mbits for the maximum bit rate and 420 kbits (α (n) x Rmin) for the minimum bit rate.

6 shows an example showing the variation of the size of the virtual buffer as a function of the bit rate for the various values of this parameter K.

The horizontal axis represents the bit rate (Mbits / s), and the vertical axis represents the size of the virtual buffer in Mbits. In curve 61, K corresponds to 0.2, and dashed line curve 62 has K of 0.12. According to the MPEG2 standard, the size of the decoder buffer is taken at 1.832 Mbits, and the bit rate variation is between 1.5 Mbits / s and 15 Mbits / s.

The curve 61 is divided into two regions, the linear region is between 1.5 Mbits / s and 9 Mbits / s, and the saturation region is between 9 Mbits / s and 15 Mbits / s.

If τ is a global bufferization duration, e.g. the elapsed time between the instant of recording the image in the encoder buffer and the instant of reading the same image from the decoder buffer, The elapsed time is similar to the passage time of the buffers as described above, and if α (n) is a duration calculated to ensure that the value τ is constant, then the equation can be:

τ.R [n] = α [n] .R [n] + Evbs [n], where τ · R [n] represents the offset of the virtual buffer as described above.

The calculation of the placement of the virtual buffer in the coder buffer is based on the relation {a (n)} given the predetermined virtual buffer size and bit rate and fixed transfer time corresponding to relation (a). b) Calculate and update continuously to meet b).

In other words, the size and position of the virtual buffer are constantly adapted to maintain a constant transition time.

Once the K value is selected, the α value is estimated at that value.

Figure 7 shows the α change (shown along the vertical axis) as a bit rate function (shown along the horizontal axis).

The α (n) value must be continuously adapted to the bit rate value to ensure that the value of τ is constant.

In the first part 71 of the curve-the region between 1.5 Mbits / s and 9 Mbits / s-i.e. in the linear region, the α value is constant and becomes:

In the second part 72 of the curve corresponding to the saturation region, the α value varies between α _max and α _min :

The minimum value of K is

The K value chosen is generally greater than this minimum value, which corresponds to no saturation region, and typically is 0.2. This is why the α (n) value must be continuously readjusted, and thus the α (n) value is not constant.

The coder according to the invention is shown in FIG. 8 and described below.

Preprocessing circuit 81 receives the data to be coded. The output of the preprocessing circuit is linked to the input side of the coding circuit 82. The output of the coding circuit is linked to the input side of the buffer 83, which supplies a binary stream to be transmitted from its output to the decoder side. The output of the buffer is also linked to the input side of the regulator 84. Since the output of the coding circuit is linked to the input side of the regulator, the regulator receives data from the coding circuit at its other input. A local supervisory facility 86 is linked to the bit rate allocator 85, preprocessing circuit 81, and regulator 84.

The preprocessing circuit 81 is included in the digital source in order to filter-shaping-and more specifically reorder the received images according to the coding type selected for each image. Receive a useful video signal in a format. Video coding circuitry 82 compresses the received signal and formats it in the form of a binary train called an "elementary stream" in MPEG2 format. This signal passes through an output buffer 83 to be sent to the decoder side in the form of a binary stream, or "bit stream". The adjuster measures the actual bit rate of the coder output from the coded data making up the binary stream sent by the buffer. The coding circuit 82 sends the coded data to the coordinator side to cause the coordinator to calculate the coding cost and the quantization time interval that returns to it. Local supervisor 86 receives the complexity factor of regulator 84 and sends it to bit rate allocator 85, which also receives such information signals from other video sources to multiplex these various video sources. By controlling the rate, the bit rate is eventually allocated to each video source. Thus, the local supervision device 86 supplies an interface with an external device and receives a phase-advanced bit rate information signal, in particular for the application of the bit rate information signal at the coder output from the bit rate allocator. It is also linked to the preprocessing circuit 81 which, among other things, sends the data of the GOP structure from the regulator 84 to the preprocessing circuit side and also causes the preprocessing circuit to reconstruct the images.

The regulator checks the quantization interval of the data compressor of the coding circuit. These regulators comprise a device that can measure the bit rate at the coder's output, as well as the decoder's VBV model buffer, which is called a virtual buffer in the coder's physical buffer, and the video decoder's buffer is neither saturated nor exhausted. Configure to ensure that

The bit rate adjuster, which affects the quantizer time interval as a function of the buffer state, can be divided into two functional sub-blocks.

The image adjustment sub-block, which performs the adjustment function only for the image level, each of the " target bit " settings corresponding to the data volume that the next image must be generated to code. Receive for the image. This item is determined by another GOP coordination sub-block that performs coordination at the GOP level as a function of configuration parameters related to the state of the coder and the VBV model buffer.

Starting from the initial quantization interval for coding the first block of the image, the image adjustment algorithm dynamically adjusts the quantization interval so that the quantization time interval is as close as possible to the programmed "target bit" value. This adjustment is performed by a feedback system in which the stiffness (convergence rate) of the feedback system can be controlled by a parameter of the same name.

At the end of the image, the image adjustment sub-block sends the actual coding cost of the image to the GOP adjustment sub-block side. The GOP transmitted coordination sub-block calculates the error between the transmitted "target bit" and the cost. The error is then fed back to the algorithm to calculate the "target bit" for the trailing image. The image adjustment sub-block also transmits the average quantization interval used to code the last image. This is used to calculate the inductive knowing complexity of the just coded image, as described below.

The principle for calculating the "target bit" in the image adjustment algorithm is as follows:

Ti, Tp (n) and Tb (m) are the target costs (ie targets) when coding an intra image, an image in a predictive image (n) and an image in a bidirectional image (m), respectively. For the GOP:

TI = target bit of type I image input,

TP = target bit of the P type image being input,

TB = target bit of type B image input,

QI = average quantization interval of type I image,

QP = average quantization interval of P type image,

QB = average quantization interval of type B images,

NP = number of P-type images in the GOP,

NB = the number of B type images in the GOP.

A GOP consisting of N images includes one intra image, NP predictive images and NB bidirectional images: N = 1 + NB + NP.

Therefore, the principle of coordination by GOP requires compliance with the following equation:

Assuming there are no coding cost changes per image type in the GOP, we can obtain the following basic equation for inter-image adjustment:

(1) T _I + N _P T _P + N _B T _B = N × B _PP -ε

(2) Z _I Q _I = Z _P Q _P = Z _B Q _B

Equation (1) expresses the principle of distributing the available bit rate for a set of images making up a GOP. However, the available bit rate is increased or decreased by an error ε corresponding to the sum of the deviations (target bit-actual coding cost) measured in the image set that constitutes the previously coded GOP.

Equation (2) establishes a rule of correspondence of average quantization intervals between different types of images in order to maintain stable quality between different types of images. The experimental values of the rate constants are as follows:

ZI = 1.0

ZP = 1.35

ZB = 1.1

The system according to equation (3) calculates the image complexity, which is the value of the coding cost of the final image of the same shape multiplied by the average quantization interval. The complexity stability assumption then makes it possible to represent the same complexity based on the target bit used to code the next image. In other words, one can make the assumption that the image complexity of the next GOP is the same as the complexity measured in the preceding GOP.

This complexity prediction makes it possible to deduce the value by deductively calculating the initial quantization interval and the target bit for each type of image:

The system according to equation (5) makes it possible in most cases to calculate the value according to the constraints in the decoder adjustment. However, errors that are shifted from one GOP to another cannot be deductively limited, so the output bit rate may be subject to some variation present in the selected run mode.

This is the reason for making the GOP coordination sub-block in the VBV model. This simplified operation of the decoder model uses the following assumptions as previously described:

The product of the coding value and the decoding value is zero;

It is instantaneous that the image is entered into the coder buffer after coding;

Retrieval of images for decoding is instantaneous in the decoder.

Thus, for each encoded image the adjustment algorithm updates the buffer occupancy state of the VBV model by applying the following relationship:

VBV_fullness (n) = VBV_fullness (n) + Bpp (n) -Last_picture_cost + εB _PP (n-τ)

As described above, the relationship indicates that Bpp bits are input into the decoder's buffer when the equivalent value of the last coded image is retrieved. The additional error signal εBpp represents the deviation between the measured bit rate and the expected bit rate at the output of the coder:

The first use of the VBV model involves padding processing (zero bytes appended to the end portion of the bit stream image) to ensure that the decoder buffer is not saturated for the current image and while transmitting the next image. The process checks whether the buffer occupancy derived from the model does not exceed the maximum size of the decoder buffer, i.e., is less than the expected data volume value (Bpp) at the input of the decoder buffer during the period of the next image. Steps. therefore:

If (VBV_fullness (n)> VBV_size-Bpp (n + 1),

Last_pict_cost (n) = Last_pict_cost (n) + VBV_fullness (n)-(VBV_size-Bpp (n + 1)),

VBV_fullness (n) = VBV_size-Bpp (n + 1)

VBV_size represents the useful size of the decoder buffer (1.835 Mbits for MPEG2 of Main Profile Main Level). The volume of padding data corresponding to the offset applied to the cost of the image is then introduced.

Besides padding, accurate perception of the VBV model buffer occupancy state allows proactively re-adjusting the target bit of the next image to be encoded. The target bit readjustment step is essential for predicting a buffer violation. Two specific cases occur:

The observed state of the VBV model buffer is too close to exhaustion. In this case, the next target bit value should be reduced by an amount that is larger the greater the risk;

The observed state of the VBV model buffer is too close to saturation. In this case, the next target bit values should be increased by an amount that increases with greater risk;

Under the assumption that the decoder's buffer is full, the buffer state initialization of the VBV model is performed. In order to ensure this condition, with respect to the principle of synchronization processing between the coder and the decoder described above, the concept of delay (end-to-end delay) when transitioning from the coder buffer to the decoder buffer is considered.

The principle of the algorithm operation is thus characterized by using the GOP adjustment sub-block to calculate the "target bit" and "initial quantization interval" parameters transmitted to the sub-block adjusting the image at the image rate.

9 schematically illustrates the various calculations performed by the GOP coordination sub-block and the sub-block coordination algorithm.

The pre-calculation module 91 receives the "coding cost" {CC (n)} and the "average quantization time interval" {Q (n)} information signals, which are previously coded images n. ) And transmitted by the image adjustment sub-block, as well as at the third input, the error {E (n)} between the expected cost {Target bit T (n)} and the actual cost {CC (n)}. Also receives. In the fourth input, the pre-calculation module is also related to the expected bit rate {Bpp (n + 1)} corresponding to the instant of coding the trailing image (n + 1) and is also a setting value starting at the bit rate allocator. Receive an information signal. As described, the bit rate allocator sends the channel bit rate setting to the coordinator before the moment τ at which the channel bit rate to be transmitted becomes effective. Thus, the allocator provides the bit rate information signal Bpp (n + 1) for the subsequent trailing image. The module 91 has target bit {T (n + 1)} information for the image n + 1 and initial quantization time for the image n + 1 at the first and second input terminals of the control module 92. Send the interval {Q_init (n + 1)}.

The bit rate information signal originating at the allocator is also sent to a delay by one image duration circuit 93. The output of this circuit is sent to the third input of control module 92. The output of the delay circuit 93 is also sent to the input on the τ period delay circuit 94 side. The output end of the τ period delay circuit is linked to the first input (+) of the first subtractor 95, which is measured at the output of the coder by the bit rate measurement circuit 96 at its second input end. Receive an information signal. The output of the subtractor is linked to the fourth input of control module 92. The control module readjusts the parameters of the virtual buffer and sends the initial target bits and quantization time interval information of the trailing image (n + 1) to its output, which is also the output of the GOP adjustment sub-block. These two information signals are sent to an image adjustment sub-block to code the next image.

The purpose of the control module is to inductively identify the risk of a buffer violation. To do so, the control module thus not only displays the difference value {(Bpp (n-τ) -Dc (n)), but also the next image, corresponding to the error value between the bit rate value set at its input and the actual bit rate value. Receive the parameters {T (n + 1), Q_init (n + 1)} and the parameters {Bpp (n)} for the image currently being coded. Calculate the variation in the fill factor.

The target bit value of the corrected trailing image {T (n + 1)} is fed back to one image period delay circuit 97, the output of which drives the input terminal (+) of the second subtractor 98. The input (-) of the subtractor receives the coding cost of the current image (CC (n)), the output of which is connected to the third input of the preceding calculation module 91.

Various calculations performed by the preceding calculation module and the control module will be described with reference to the flowchart shown in FIG. 10.

The first step 101 is an interrupt wait loop for triggering a task. The image interrupt informs the algorithm that the information presented in the image related to the image n just encoded is available. This step is followed by a step 102 of reading the following information, where the information read is as follows:

Coding cost of image n {CC (n)},

Mean quantization time interval for image n ,

Image-related bit rate settings for image (n + 1) {Bpp (n + 1)},

-Type (I, P, B) of image (n + 1),

The bit rate {Dc (n)} at the coder output, for image n.

Step 103 then calculates a "target bit" for the image according to each type. The result of coding the preceding image (coding cost and average quantization time interval) is used to calculate the complexity of the image n.

The error {E (n)} for the "target bit", which occurs when coding, is calculated for each of the images n and is stored in the end of the GOP via the GOP to be stored as a value ε. Accumulate. The E (n) value is then initialized in the first image of the GOP.

E (n) = E (n-1) + T (n)-CC (n)

The new "target bit" is calculated as follows for each type of image (n + 1):

Therefore, these "target bits" consider the accumulation error of the preceding GOP. This error is then redistributed across the trailing GOP images to compute the target bit.

Step 104 then performs an inductive check for the risk of virtual buffer violations, readjusts the "target bit" and calculates the initial quantization interval for the next image to be coded.

Thus, the step executes updating the VBV model by calculating the VBV_fullness state. The buffer state makes it possible to insert padding first to avoid saturation of the decoder's buffer. Secondly, the 'target bit' may possibly be reconditioned (the reconditioning process is not the subject of this patent), and the initial quantization interval of the next image to be coded is calculated.

E _C (n) = B _PP (n-τ)-D _C (n)

VBV_fullness (n) = VBV_fullness (n-1) + Bpp (n)-[CC (n) + B (n)]-E _C (n)

CC (n) + B (n) is an image of an image added with a padding volume {B (n)} corresponding to the number of bits (for example, 0s) input to the coder's buffer at the same time that the image n is input. Cost {CC (n)}.

If VBV_fullness (n) is greater than VBV_size-B _PP (n + 1), the padding volume {B (n)} is modified as follows:

B (n) = VBV_fullness (n)-(VBV_size-Bpp (n + 1))

The variation in the fill factor of the buffer is:

VBV_fullness (n) = VBV_fullness (n) -B (n)

The initial quantization time interval is calculated as follows:

QIinit = XI / TI

QPinit = XP / TP

QBinit = XB / TB

Step 104 then returns to initial step 101.

Note that the concept of virtual buffer and VBV_offset is implicitly related to the fact that the measured bitrate value at the coder output at instant n is compared with the value Bpp at instant n-τ. do. The offset of the virtual buffer is then adjusted by feeding back the mismatch between these two items in calculating the state of VBV_fullness. This is a simple feedback loop that ensures position control of the virtual buffer.

Applications relate to digital television, in which the transmission of groups of programs, ie the simultaneous transmission of several video sources, such as the satellite communication of video sources or the sharing of transponder channels, is achieved by means of statistical multiplexing. use. Video data is compressed and multiplexed so that it can be transmitted packwise over a channel.

The application also relates to DVD recording using multipass encoding. The first pass allows to determine the complexity of the images in the entire film. The second pass makes it possible to assign a bit rate to each image according to the correlation with the complexity starting from the average bit rate described above. Thus, the quality of the images is constant for the same volume of data.

As described above, the present invention is used in a method of coding audio digital data, video digital data, or auxiliary digital data, and more particularly, the present invention uses a compression scheme for audio and video digital data, and thus, a variable bit rate. Is used in systems that transmit and store data.

Claims (11)

The coder to perform compression of the data image, adjust the bit rate at the output side of the coder using an adjustment loop that affects the quantization interval of the data to be coded, and transmit the data through the channel to the buffer side of the decoder at a variable bit rate Storing the data output by 83 in a buffer and ensuring a constant time τ between the instant of storing the image in the buffer of the coder and the instant of outputting the image from the buffer of the decoder. Tagging and bit rate adjustment on the image at instant n is performed as a function of bit rate prediction of the transmission channel with respect to instant n + τ, where the fullness of the coder buffer defines the virtual buffer. Moving in time between the two limits, the upper limit 46, or virtual buffer offset, calculated for coding the image at instant n is small. A method of coding digital data of a video image sequence, which is proportional to the transmission bit rate of a channel predicted for an instant in time (n + τ) over a range of variation in the bit rate. The size of the virtual buffer at the instant n, defined between the two limits, is proportional to the transmission bit rate of the channel predicted for the instant n + τ over at least the fluctuation range of the bit rate. A method of coding digital data in a video image sequence. The method of claim 1, The size of the virtual buffer is calculated to be proportional to the channel transmission bit rate over at least the variation range of the bit rate, and the change from the size corresponding to the previous bit rate to the size corresponding to the new bit rate is time for the transmission bit rate to change. A method of coding digital data of a video image sequence, characterized in that it occurs continuously from before (τ). 3. The method according to claim 1 or 2, wherein the transmission bit rate predicted at the instantaneous time point (n + τ) is the average complexity of the image that was before the image during the duration of a group of pictures (GOP). Characterized in that it is calculated as a function. The method of claim 1, wherein the modeling of the virtual buffer is performed by coding each image in a manner of supplying a bit rate at the buffer input side, which corresponds to a bit rate allocated to the output of the buffer after the instant τ. A method of coding digital data of a video image sequence. 5. The digital data of a video image sequence according to claim 4, characterized in that the model of the virtual buffer is inductively updated by merging the error between the output bit rate and the bit rate assignment allocated before the instant τ. How to code it. 2. The method of claim 1, wherein the maximum size of the virtual buffer is the size of the decoder buffer. The video image sequence of claim 1, wherein an upper limit value and a lower limit value of the bit rate variation range at the output side of the coder correspond to the virtual buffer position at an upper boundary value and a lower boundary value of a physical buffer. How to code digital data. 2. The method of claim 1, wherein the adjustment of the image n + 1 is performed based on an initial quantization interval and the number of target bits or targeted bits for the coding, wherein the target bits are image periods n. calculated from a setting of an expected channel bit rate at the coder output side for + τ). The digital image of a sequence of video images according to claim 8, characterized by considering an image coding type which is intra coding, predictive coding, bidirectional coding and a predetermined proportional constant value to calculate the target bit value. How to code your data. 9. The method of claim 8, wherein coding of an image belonging to a GOP at instant n is performed by merging an error between the target bit calculated for the image and the coding cost of the image, wherein the error is a previous GOP. A method for coding digital data of a video image sequence, characterized in that it is accumulated at a level. A method of transmitting data from several video sources, the method of performing the data coding of claim 1 for each source, in order to transmit data at a variable bit rate over a predetermined transmission channel. In the transmission method, According to the useful bit rate of the channel and the complexity measurement of previous images from the set of video sources, for each source a bit rate assignment is performed before the instant τ of the source's actual bit rate over the channel. , Data transfer method.