JPH0799810B2 - Encoding output data amount control system and decoding device thereof - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial field of use) In a recording, transmission, and display device for processing a digital signal, an efficiency coding method for efficiently coding a signal with a smaller code amount is output. The present invention relates to a coded output data amount control method for making the amount of data to be reproduced uniform.

(Prior Art) Prediction residuals in predictive coding and orthogonal transform coefficients in exchange coding have a considerable bias in the level distribution of generated signals.

Therefore, by assigning codes of different lengths according to the frequency of occurrence of the signal level, the average value of the data amount can be shortened with respect to the fixed-length code.

The variable length code is determined by a Huffman code or the like, and the more the data generation frequency is biased, the lower the total data amount can be.

By using the variable-length code in this way, the average data amount can be made smaller than that of the fixed-length code while maintaining the quality of the reproduced signal equal.

However, since the amount of generated data is not constant, it is necessary to control the amount of generated data to be uniform in an actual device.

Feedback control by a data output buffer is one of the most common control methods for controlling the amount of data.

When the amount of encoded data increases and the amount of data stored in the buffer increases, the quantization step is roughened to reduce the amount of data.

Such a control system is used in a system such as a video conference.

On the other hand, when considering the application of encoding to a recording medium, not only simple recording and reproduction but also functions such as special reproduction are desired.

Further, in the editing of encoded image data, the data length needs to be fixed in units of about one frame in order to replace the image data.

In such a case, although the average value of the data amount is kept constant by the feedback control, the data amount cannot be surely kept within the constant value in a desired unit.

Therefore, as a data amount control method for keeping the data amount within a fixed value, there is a prior application by the same applicant and the same inventor of the present application, Japanese Patent Application No. 1-14273, "Encoding output data amount control method". is there.

This method uses both variable-length coding and controls both feedforward and feedback to ensure that the amount of data is kept within a certain value without deterioration of coding efficiency or deterioration of image quality. .

(Problems to be Solved by the Invention) In conventional feedback control, high coding efficiency can be obtained by variable-length coding, but the amount of data cannot be reliably stored within a fixed value in units of frames, etc. There was a problem that it was difficult to apply to media.

On the other hand, the Japanese Patent Application No. 1-14273, "Control system of encoded output data amount", which controls high efficiency encoding for normalization, controls the data amount per block by a normalization coefficient. , And the decoding side uses the transmitted normalization coefficient to control the quantization by the same processing as the encoding side, so it is assumed that the normalization processing has been performed, and normalization is not performed. There is a problem that it cannot be applied in the case of encoding and has no versatility.

The present invention has been made by paying attention to the above points, and controls the amount of data by both feedforward and feedback, and the control information of how the feedback control is performed is the variable length code of the orthogonal transform coefficient. Since it is encoded in the same way as the encoding and transmitted to the decoding side, the amount of data can be reliably stored within a fixed value in frame units, etc., and can be applied to storage media, and special normalization processing etc. are not required. It is an object of the present invention to provide a versatile encoded output data amount control system and a decoding apparatus therefor.

(Means for Solving the Problems) In order to solve the above problems, the present invention (1) controls the data amount so that the encoded output data amount for each predetermined constant section is within a predetermined value. In a coded output data amount control method for performing high-efficiency coding by predicting the data amount in units smaller than the fixed section, the total of the predicted data amount in the fixed section becomes constant. Such a predicted data amount, a feedforward control means for setting the value of the temporary quantization step, and a difference between the predicted data amount and the actually encoded actual data amount are accumulated, and with this accumulated data amount, If the data amount is larger than the predicted data amount, the final quantization step becomes coarse, and if the actual data is smaller than the predicted data amount, the final quantization step becomes finer. A feedback control means for correcting the value of the quantization step, and control information coding means for coding the control information for the quantization step controlled by the feedforward and feedback control means and transmitting it to the decoding side. And a control information decoding means for decoding the control information encoded and transmitted using the coded output data amount control method according to claim (1). The present invention provides a decoding device of an encoded output data amount control system, characterized in that the decoding device is constituted by an encoded data decoding unit that performs an operation of decoding encoded data under the control of the decoded control information. .

(Operation) In order to solve the above problem, the data amount is controlled by both feedforward and feedback, and control information by feedback is variable length coded and transmitted to the decoding side.

To control the amount of data, first, the predicted value of the amount of data is obtained by the activity of each block, and the provisional quantization step is set by feedforward control. Here, as the activity, the root mean square value of the AC components of the orthogonal transformation coefficient or the like is used.

The difference between the data amount and the predicted value that occurs in the actual variable length coding is set so that the error accumulation does not become large by shifting the quantization step in the feedback control in block units.

The control information on how the feedback control is performed is encoded by the same method as the variable length encoding of the orthogonal exchange coefficient and transmitted to the decoding side.

First, the feedforward control makes the data amount approximately the same as the target data amount. Further, the error of the data amount can be suppressed to about several blocks by the feedback processing in block units. The fluctuation range of the data amount is a loss for the target data amount, but it is extremely small.

Since the feedback control is performed with respect to the prediction error of the data amount, the image quality does not greatly change even in the block unit control.

Since the feedback control information is encoded and transmitted,
Although the data amount increases by that amount, it is only a little compared with the original data amount of encoding.

As a result, the amount of data is fixed in a desired unit while performing variable length coding.

(Embodiment) FIG. 1 is a block diagram showing an embodiment of a coding device of a coded output data amount control system of the present invention.

The image input signal is input from the image input terminal 1, supplied to the high-efficiency encoding processing unit 2, and subjected to high-efficiency encoding processing,
An image output signal is taken out from the encoded data output terminal 3.

The high-efficiency encoding processing unit 2 further includes an orthogonal transformer 4, a coefficient memory 5, a quantizer 6, a variable length encoder 7, and a variable length buffer 8.

This high-efficiency encoding processing unit 2 is an intra-frame processing such as CCITT of the International Telegraph and Telephone Consultative Committee, SG15 standardized video conferencing system (p × 64 kbps codec), etc., except that a coefficient memory 5 is added. It is the same as the section.

It can be said that this is general and versatile in that the processing unit of Japanese Patent Application No. 1-14273 of the prior application requires the coefficient normalization, but does not require it.

The operation of the high efficiency encoding processing unit 2 will be described below.

The image input signal is input from the image input terminal 1 and supplied to the orthogonal transformer 4, and is first converted into a coefficient for each block such as 8 × 8 pixels.

As well known, the discrete cosine transform (DCT) is used as the orthogonal transform method, and one of the transform coefficients is the DC coefficient (average).
And all other conversion factors are AC factors (AC).

Each coefficient which is the output signal of the orthogonal transformer 4 is supplied to the coefficient memory 5.

The coefficient memory 5 is for compensating for the delay by the feedforward control described later, and is a memory for a section (here, for example, one frame) where the data amount is desired to be constant.

Since the number of coefficients is equal to the number of pixels before conversion, the coefficient memory 5
Is similar to the frame memory, but the accuracy of the converted value is required to be higher than that before the replacement. For example, 8 × 8DCT per coefficient
It requires about 12 bits.

The transform component that is the output signal of the coefficient memory 5 is the quantizer 6
And is equally quantized at intervals of a quantization step Sq described later.

The output signal of the quantizer 6 is supplied to the variable length encoder 7.

Here, most of the AC components are concentrated in the vicinity of 0, so they are converted to a preset Huffman code or the like. In this Huffman code, the word length of 0 is usually the shortest, and the word length becomes longer as the absolute value of the level becomes larger.

Furthermore, 0 is not coded for each coefficient, but the number of 0s is coded.

The amount of output data thus obtained differs from block to block.

The output signal of the variable length encoder 7 is supplied to the variable length buffer 8.

The variable length buffer 8 is a FIFO (First In First Out) type memory, and although the input speed changes according to the amount of data output from the variable length encoder 7, it is temporarily stored and converted to a constant data rate. The high-efficiency coded image coded data is taken out via the coded data output terminal 3.

Next, of the control system of the data amount which is the characteristic part of the present invention,
First, the feedforward control unit 9 will be described.

The feedforward control unit 9 further includes an activity detector 10, an activity memory 11, and a quantization step temporary setting device 12.

The feedforward control unit 9 outputs a temporary quantization step PSq determined for each frame based on the activity (activity) of the orthogonal transform coefficient. The operation will be described below.

First, each coefficient which is the output signal of the orthogonal transformer 4 is supplied to the activity detector 10.

The activity detector 10 obtains the activity (activity) for each block from the conversion coefficient. This block does not necessarily have to be the same as the orthogonal transform block, and some orthogonal transform blocks may be combined.

Here, if this block is made large, the data amount of the feedback control information, which will be described later, becomes small, but the fluctuation of the data amount also becomes large, resulting in a loss. Therefore, it is desirable that the two be balanced.

Since the activity is used for predicting the data amount of variable length coding, the higher the correlation, the better, and the activity is given by the root mean square value of AC coefficients. This value is not directly related to decoding, so if a better activity detection method were later developed, it would not cause any compatibility issues.

The requested activity is the activity memory 11
And is supplied to the quantization step temporary setting device 12.

The activity memory 11 performs an operation of delaying by one frame to compensate for the delay of the feedforward control, and since the activity is for each block, the activity memory 11 can have a small capacity as compared with the coefficient memory 5 and the like.

The quantization step provisional setting device 12 operates to obtain the provisional quantization step PSq such that the data amount of one frame becomes constant.

FIG. 2 is a block diagram showing a configuration example of the quantization step provisional setting device in FIG. 1, and particularly shows a configuration example in the case of parallel processing.

In FIG. 2, a plurality of data amount predictors, frame accumulators, and comparators are provided for each quantization step value, but the data amount predictor is different for each quantization step, and the other processing is the same. Is.

The quantization step is set in advance by changing the value by about 10% and n kinds (for example, n is about 20).

From the activity detector 10, the activity input when quantized in each quantization step is supplied to each data amount predictor 16, 17, 18, etc. via the activity input terminal 15, and the predicted data amount PDb for each input activity
1, PDb2, ... PDbn are output.

This predicted value is an average value statistically obtained in advance, and the predicted data amount increases as the quantization step becomes finer. The values are written in the table ROM, and the predicted data amount is output as soon as the activity is input.

In this way, the predicted data amount PDb obtained in block units
1, PDb2, ... PDbn are accumulated for one frame by the frame accumulators 19, 20, 21 to obtain the predicted data amount D1, D2, Dn for one frame at each quantization step.

D1, D2, ... Dn are compared with the target data amount by the data amount comparators 22, 23, 24 at the time when the frame ends at each quantization step, and it is determined whether the data amount is smaller or larger than the target. .

The result is input to the appropriate value determiner 25, which determines the finest quantization step among the smaller amount of target data, and outputs the quantization step as the temporary quantization step PSq via the output terminal 26. is doing.

Since each input of the appropriate value determiner 25 is 1-bit information, the circuit can be constructed by simple logic.

Next, the operation of the feedback control system 28 in FIG. 1 will be described.

From the activity delayed in the activity memory 11 during the feedforward control and the temporary quantization step PSq output from the quantization step temporary setting device 12, the data amount predictor 29 obtains the predicted data amount PDb of each block.

The data amount predictor 29 is the same as the data amount predictor 16, 17, 18 inside the quantization step temporary setting device shown in FIG. 2, but the contents of the ROM are the same as those of the data amount predictor 16, 17, 18. Only the same as all are put together.

The predicted data amount PDb is subtracted by the subtractor 30 from the actual data amount Db given from the variable length encoder 7, and is guided to the data amount accumulator 31.

The data amount accumulator 31 adds (Db-PDb) to the cumulative value K up to that point and holds a new cumulative value K for each block. The cumulative value K is reset every frame.

The accumulated value K output from the data amount accumulator 31 is supplied to the quantization step shifter 32, and the quantization step shift amount DS
converted to q.

FIG. 3 is a diagram showing a quantization step shift characteristic. The horizontal axis is the cumulative value K of the data amount error, and the vertical axis is the quantization step shift amount DSq.

In FIG. 3, the maximum cumulative value K is Kmax and the minimum cumulative value is −
Although Kmax is used, this Kmax is a margin that must be finally set for feedback control.

The characteristic of FIG. 3 is that the shift amount DSq increases toward the end so that K does not exceed this value in feedback control.

The final quantization step Sq is obtained by adding the shift amount DSq and the provisionally set temporary quantization step PSq in the adder 33.

Since a statistical average value is given to the predicted data amount PDb, in a general image, the integrated value of the predicted data amount PDb and the integrated value of the actual data amount Db are equal, and K is close to 0. Becomes

However, if the actual data amount Db occurs unevenly,
K has a positive or negative value, and the quantization step Sq shifts with respect to the temporary quantization step PSq.

As described above, the feedback control in the present invention is performed for the prediction error of the data amount and is finely controlled in block units.

If the control is properly performed here, the error in the data amount of one frame at the end of one frame is within Kmax, and the temporary quantization step PSq set in the feedforward system is set.
The estimated data amount in suffices to have a margin of Kmax with respect to the target data amount.

When Kmax is the data amount for 10 blocks, the amount is about 1/540 of the total if the number of pixels in one frame is 720 × 480 and the block size is 8 × 8, and the decrease in encoding efficiency is almost ignored. it can.

On the other hand, such a control circuit is relatively complicated, but since the processing other than the activity detector 10 is performed in each block, the processing intensity is about 1/64 of the original signal, and a general-purpose DSP (digital signal processor) is required. ) Is also possible.

Next, the control information coding means will be described.

The control information that needs to be transmitted to the demultiplexing side is the shift amount of the quantization step that is the result of the feedback control. Since the quantization step is set about every 10%, the change of the quantization step shift amount DSq does not occur for each block, and it is considered to be once for every several blocks.

Therefore, the data amount can be significantly reduced by performing run-length coding on the part where there is no change.

FIG. 4 is a diagram showing a variation example of the quantization step shift amount. The horizontal axis represents the change interval (that is, the unit of time), and the vertical axis represents the quantization step shift amount DSq.

For example, if the control shown in FIG. 4 is performed, the information that needs to be transmitted is the change interval and the change degree.

The change over one degree is almost ± 1 step.
(In FIG. 4, since 1 step is 2%, ± 2
% Is changing. ) For example, in FIG. 4, {6, +2} indicates that the change interval is 6 and the change degree is + 2%.

Therefore, the change interval is used as the run length, and the Huffman code or the like is used together with the degree of change. This technique is similar to the encoding of orthogonal transform. The processing is performed by the variable length encoder 34, and the encoded control information is output through the control data output terminal 35 and transmitted to the decoding side (decoding device).

Assuming that the block size is 8 × 8 pixels and the change in the quantization step is once every 10 blocks on average, the data of one change is about 6 bits, and the data amount is about 0.01 bit per pixel.

Since the normal data amount such as conversion coding is 1 to 2 bits per pixel, it can be said that the data amount of this control information is sufficiently small.

Although it is necessary to send the information of the temporary quantization step PSq which is the result of the feedforward process, the number of PSq is about 20, and the data amount is negligible because it is several bits per frame.

FIG. 5 is a block diagram showing an embodiment of the decoding device of the encoded output data amount control system of the present invention.

In FIG. 5, the basic operation of the decoding means for the encoded data transmitted from the encoding device is general.

First, since the pixel signal (encoded data input signal) input from the encoded data input terminal 37 has a different data amount for each block, it enters the variable length buffer 38 and is extracted for each block.

Next, the variable length decoder 39 converts the data into a fixed length, which becomes data for each coefficient.

Each coefficient data is replaced by a representative value of each quantization in the quantization regenerator 40 and supplied to the inverse orthogonal transformer 41.

The inverse orthogonal transformer 41 performs the reverse processing of the orthogonal transformer 4 in FIG. 1 to obtain a reproduced image.

The obtained reproduced image is output via the reproduced image output terminal 42.

Next, the control information restoring means, which is a feature of the present invention, will be described.

The present invention is characterized in that the quantization reproduction information (inverse quantization) is controlled by the transmission of the quantization control information for each block.

The control information (control data) input from the control data input terminal 43 is decoded by the variable length decoder 44, and the quantization step shift amount DSq for each block is obtained.

DSq is added by the adder 45 to the provisional quantization step PSq transmitted in advance at the beginning of the frame, and becomes the actual quantization step Sq.

The quantization step Sp is supplied to the quantization regenerator 40,
Thereby, the operation of quantization reproduction is controlled, and the reproduction value is obtained.

By such an operation, a correct reproduction value can be obtained even if the quantization is controlled for each block.

Further, in this method, since the control information of the quantization step is transmitted, the control information decoding means is required, but it is not necessary to have the feedback control unit in the decoding system (decoding device) as in the prior application.

As described above, in the present invention, by controlling the data amount by both feedforward and feedback, and transmitting the control information by variable length coding, in general high efficiency coding using variable length coding, The data amount can be made constant with extremely high accuracy in a target section without deterioration of image quality and deterioration of coding efficiency.

Since the rough control is performed by the feedforward control, the image quality does not largely change by the feedback control in block units, and the capacity of the variable length buffer for data output may be small.

Since the method of predicting the data amount can be performed by the activity of the block, more accurate feedforward control can be performed. Even if the activity calculation method is changed later, the compatibility problem does not occur.

Since the main processing is performed in block units, the processing speed may be quite slow, and it can also be realized by a general-purpose DSP or the like.

Due to such characteristics, highly efficient encoding processing using variable-length encoding can be applied to storage media that require special reproduction or replacement of data in frame units.

(Effects of the Invention) The encoded output data amount control method and the decoding device thereof according to the present invention control the data amount by both feedforward and feedback, and the control information indicating how the feedback control is performed is the orthogonal transform coefficient. Since it is encoded and transmitted to the decoding side in the same way as the variable length encoding of, the amount of data can be reliably stored within a fixed value in frame units, etc., and it can be applied to storage media and special normalization There is no need for treatment, etc., and it is extremely versatile and practically excellent.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a coding device of a coded output data amount control system of the present invention, FIG. 2 is a block diagram showing a configuration example of a quantization step provisional setting device in FIG. 1,
FIG. 3 is a diagram showing a quantization step shift characteristic, FIG. 4 is a diagram showing a variation example of the quantization step shift amount, and FIG. 5 is an embodiment of a decoding device of the encoded output data amount control system of the present invention. It is a block diagram shown. 1 ... Pixel input terminal, 2 ... High efficiency coding processing unit, 3 ...
... coded data output terminal, 4 ... orthogonal converter, 5 ... coefficient memory, 6 ... quantizer, 7,34 ... variable length encoder, 8,
38: Variable length buffer, 9: Feedforward controller, 10: Activity detector, 11: Activity memory, 21: Quantization step temporary setting device, 15: Activity input terminal, 16, 17, 18 , 29 …… Data amount predictor, 19,20,21 …… Frame accumulator, 22,23,24 …… Comparator, 25 …… Appropriate value judger, 26 …… Output terminal, 28 …… Feedback controller , 30 …… Subtractor, 31 …… Data amount accumulator, 32 …… Quantization step shifter, 33 …… Adder, 35, 4
5 ... Control data output terminal, 37 ... Encoded data input terminal, 39,44 ... Variable length decoder, 40 ... Quantization regenerator, 41
...... Inverse orthogonal transformer, 42 ...... Playback image output terminal, 43 …… Control data input terminal, D1 to Dn …… Predicted data length for 1 frame, Db …… Actual data amount, DSq …… Quantization step Shift amount, K ... cumulative value, Kmax ... maximum value of cumulative value K,
PDb, PDb1 to PDbn ... predicted data amount, PSq ... temporary quantization step.

Claims (2)

[Claims] 1. A coded output data amount control system for performing high-efficiency coding by controlling the data amount so that the coded output data amount for each predetermined constant section is within a constant value. A feedforward that predicts a data amount in units smaller than the fixed interval and sets a predicted data amount and a value of a temporary quantization step such that the total of the predicted data amounts in the fixed interval becomes constant. The control means accumulates the difference between the predicted data amount and the actually encoded actual data amount, and if the actual data amount is larger than the predicted data amount with this accumulated data amount, the final quantization step is rough. So that when the actual data amount is smaller than the predicted data amount, feedback control means for correcting the value of the temporary quantization step so that the final quantization step becomes finer, Over-forward and encodes control information about the controlled quantization step by the feedback control means, the encoded output data amount control method characterized in that it is composed of a control information encoding means for transmitting to the decoding side. 2. Control information decoding means for decoding control information coded and transmitted using the coded output data amount control system according to claim 1, and controlled by the decoded control information, A decoding device of a coded output data amount control system, characterized by comprising coded data decoding means for performing a decoding operation of coded data.