JP3224666B2 - Data compression device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

FIELD OF THE INVENTION The present invention relates to a video system, tele
The present invention relates to a data compression device used for video broadcasting, video conference, videophone, etc., and compressing data such as image data.

[0002]

2. Description of the Related Art In recent years, as a compression method for still image data, a method combining discrete cosine transform (DCT) and variable length coding is being internationally standardized by the International Standardization Group (JPEG).

In this method, a discrete cosine transform is applied to still image data, and the image data is subjected to a frequency coefficient value (DCT).
Coefficient), and then quantize (specifically, divide) the DCT coefficient by a predetermined quantization constant (quantization parameter) Q to eliminate data redundancy, and then apply Huffman coding or the like. The data is encoded and recorded in a compressed form on a recording device such as an optical disk. Here, the quantization parameter Q is defined, for example, by the following equation. u is an appropriate constant.

[0004]

## EQU1 ## Quantization matrix (i, j) = (Q / u) × default visibility table (i, j)

[0005] That is, the quantization parameter Q
Default visibility table with function examples presented in G
Default visibility table when (i, j)
By changing the level of (i, j) linearly, it functions as a scaling factor (compression rate control factor) for obtaining a scaled (compression rate controlled) quantization matrix (i, j). Accordingly, such a data compression process strictly quantizes the DCT coefficients obtained as a result of performing discrete cosine transform on image data by using a quantization matrix (i, j). For convenience of explanation, data
It is assumed that (image data) is to be quantized by a quantization parameter Q.

In recent years, attention has been paid to applying this type of image data compression technology to data compression of pseudo moving images. A problem in such a case is the occurrence of a so-called overflow in which the data amount as a result of compression exceeds a prescribed limit data amount. FIG. 4 is a diagram showing a general relationship between a quantization parameter Q and a data amount D obtained as a result of quantization (compression) by using various image data α, β, γ, and δ. Referring to FIG. 4 , a QD curve representing the relationship between the quantization parameter Q and the data amount D is as follows:
It is exponential, and as the quantization parameter Q increases, the data amount (code length) D decreases (the compression ratio increases) and the image quality deteriorates. On the other hand, the quantization parameter Q
Is smaller, the data amount (code length) D is larger (the compression ratio is smaller). As shown in FIG. 5, when recording on a recording device, if this data amount D exceeds the limit data amount _Dmax , For example, the data cannot be stored in, for example, one record of the recording apparatus, and an overflow occurs.

If an overflow occurs,
For example, discard the data when the overflow occurred,
Instead of this, measures such as recording the data at the previous time point can be considered, but at this time, at the time of reproduction, the reproduced image (reproduced moving image) is momentarily stopped or disturbed, making it very difficult to see . Therefore, in order to reliably avoid the occurrence of overflow and to minimize the deterioration of image quality, a data amount somewhat smaller than the maximum data amount D _max is set as the target data amount (optimal data amount) D _obj. It is preferable to determine the quantization parameter Q so that the compression result has a data amount as close as possible to the target data amount _Dobj .

For this reason, conventionally, for example, Japanese Patent Laid-Open No.
As shown in US Pat. No. 8,873, a system using feedback control has been proposed.

[0009]

[SUMMARY OF THE INVENTION However, in the conventional method described above, to converge the data amount target data amount D _obj for one feedback control is sufficiently focused to be flame Kashiku, the target data quantity D _obj In general, in order to converge to an optimum data amount that does not cause overflow, feedback control must be repeatedly performed a plurality of times. As a result, there is a problem that it takes a considerable time to determine the quantization parameter Q, and the compression process cannot be performed in real time.

In particular, even when the scene changes continuously, if the quantization parameter Q is small, it is not possible to converge to an optimum data amount, and there is a problem that an overflow occurs. FIG. 6 is a diagram showing a QD curve relating to two pieces of image data when there is continuity between two consecutive frames (fields) and the scene changes continuously. In FIG. 6, the n-th frame
The Q-D curves for the image data of the (field) and P _n, when the (n + 1) th Q-D curves for the image data of a frame (field) and P _{n + 1,} a large place of the quantization parameter Q is Since the slopes of the QD curves P _n and P _{n + 1} are gentle, and therefore, even if the quantization parameter Q slightly changes, the data amount D does not change much. Frame
From the information on the (field) image data, a quantization parameter for the (n + 1) th frame (field) image data is predicted, and this quantization parameter is used.
Even if the image data of the (n + 1) th frame (field) is compressed, the overflow problem does not occur much.

However, as can be seen from FIG. 6, the data amount D sharply increases as the value of the quantization parameter Q decreases (because the slope of the graph becomes steep), so that the value of the quantization parameter Q decreases. In a small range, even a small change in the quantization parameter Q greatly changes the amount of compressed data.

That is, when there is continuity between two consecutive frames (fields) and the scene changes continuously, the image data of the (n + 1) th frame (field) is changed to the nth frame (field). ((N)
The QD curve P _{n + 1} relating to the image data of the (+1) th frame (field) is the nth frame (field)
(Even if the difference does not change much from the QD curve _Pn for the image data of (n)), when the quantization parameter Q becomes small, the compressed data amount of the image data of the (n + 1) th frame (field) and the nth frame The difference between the (field) image data and the compressed data amount becomes large.

In the conventional data compression system based on feedback control, the n-th frame (field) is used.
, The quantization parameter Q for the (n + 1) th frame (field)
To predict the _{n + 1,} with respect to a slight change of the quantization parameter Q, only if the data amount does not change significantly not assumed, only the change of the quantization parameter Q in a small range of the quantization parameter Q If the amount of compressed data greatly changes, the predicted quantization parameter Q _{n + 1}
When the quantization is performed on the image data of the (n + 1) -th frame (field) using, the data amount easily exceeds the limit data amount _Dmax , and overflow easily occurs. As a countermeasure at this time, as described above, it is conceivable that the data at the time of overflow is discarded and the data at the previous time is recorded instead. When the data recorded in such a manner changes, the reproduced image (reproduced moving image) is momentarily stopped or disturbed, making it very difficult to view the image.

According to the present invention, even when the compression ratio (quantization parameter) becomes small, overflow can be effectively prevented, data quality degradation can be suppressed, and data compression capable of performing compression processing in real time can be performed. It is intended to provide a device.

[0015]

To achieve SUMMARY and operation for solving the above objects, the invention of claim 1 Symbol placement on the basis of the compression result obtained by performing compression processing on the data at some point, the next Quantization parameters for the data at
A data compression apparatus for compressing data by a feedback control to predict the data, against the data in the next time
The quantization parameter for the data at a certain point in time
Compressed parameters and the result of compressing the data at some point
And the limit data amount at which overflow occurs
And the quantization parameter predicted by the feedback control.
When the quantization parameter becomes smaller than the predetermined quantization parameter, the quantization parameter for the data at the next point in time is
It is determined to the determined quantization parameter ,
In this case, the predetermined quantization parameter is
Data when compression is performed with the specified quantization parameter.
The fluctuation range of the data amount from the target data amount is the limit data amount
Is set to a range that can be absorbed by the difference between the data amount and the target data amount . As a result, even when the quantization parameter is reduced, the data amount can be converged to a data amount as large as possible without causing overflow, thereby suppressing the deterioration of image data quality and performing the compression processing in real time. .

[0016]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a configuration diagram of an embodiment of a data compression device according to the present invention. Referring to FIG. 1, the data compression apparatus converts image data of an n-th (n: integer) frame (field) into a predetermined quantization constant (quantization parameter) Q _n.
An image compression unit 2 for compressing by compressing the image data, a buffer memory 22 for temporarily storing the compressed image data, a recording device 23 for recording the compressed image data, A data amount detection unit 4 for detecting a data amount D _n obtained by compressing the image data of the frame (field), and a compression ratio for the image data of the (n + 1) th frame (field), that is, a quantization parameter Q a compression ratio control unit 5 for predicting _{n + 1} ′ using the quantization parameter Q _n , data amount D _n, and target data amount D _obj used for the image data of the n-th frame (field); The compression ratio for the (n + 1) th image data, that is, the quantization parameter Qn _{+ 1} is finally determined based on the quantization parameter Qn _{+ 1} 'predicted by the compression ratio control unit 5. And a compression ratio determining unit 16.

Here, the data amount detection unit 4 is constituted by, for example, a counter. In addition, the target data amount D
_obj is set to a value slightly smaller than a limit (maximum) data amount _{Dmax at} which an overflow occurs in order to avoid an overflow due to a change in the data amount due to a calculation error. Further, the compression ratio determining unit 16 sets a predetermined limit value (minimum value) Q _min for the quantization parameter Q, and sets a range (1) where the predicted quantization parameter Q _{n + 1} ′ is smaller than Q _min. When <Qn _{+ 1} ≦ _Qmin ), the quantization parameter Qn _{+ 1} is determined to be a constant value, that is, a limit value _Qmin .

The above-mentioned limit value (minimum value) Q _min of the quantization parameter Q is in a range where the value of the quantization parameter does not become too small, that is, in a case where the scene changes continuously, the predicted quantum The variation range of the data amount ( the variation range from the target data amount _Dobj ) when the compression is performed by the optimization parameter is set to the limit (maximum) data amount D
It is preferable to set a range that can be absorbed by the difference (margin) between _max and the target data amount _Dobj .

Next, an example of processing of the data compression apparatus of FIG. 1 will be described with reference to FIG. In order to easily and quickly perform the data compression processing of the present invention using a microprocessor or the like, an exponential QD curve relating to image data is approximated to, for example, the following equation.

[0020]

## EQU2 ## Q = a / D ² + b

Here, a and b are constants peculiar to the image and differ depending on the type of the image.
There is almost no difference in the numerical value depending on the type .

Also, the data compression apparatus of the present embodiment assumes that the scene changes continuously, so that the (n + 1) -th image hardly changes from the n-th image, and (n +
1) The QD curve P _{n + 1} of the n-th image is
It is assumed that the operation is performed under the assumption that the D curve _Pn is the same (the constants a and b unique to the image are the same).

[0023] In this case, the image compression unit 2 performs quantization by the quantization parameter Q _n for n-th image, the relationship between the resultant data amount D _n and Q _n, the following equation become.

[0024]

[Number 3] ^{_{Q n = a / D n 2}} + b

As the quantization parameter Q _{n + 1} ′ predicted by the compression ratio control unit 5 for the (n + 1) -th image data, the quantization parameter Q _{n + 1} ′ is It is best to make the data amount of the converted and compressed result the target data amount _Dobj .

[0026]

## EQU4 ## Q _{n + 1} '= a / D _obj ² + b

[0027] Now, for example, in the case of a large amount of information scene, the quantization parameter Q takes a large value, as can be seen from Figure 2, the range AR ₁ quantization parameter Q takes a large value, Q-D curve Is gentle. Accordingly, the (n + 1) -th image data slightly changes from the n-th image data (in FIG. 2, the QD curve P _{n + 1} for the (n + 1) -th image data shows the QD curve for the n-th image data. (Slightly changes from the curve P _n ) and the quantization parameter Q _{n + 1} ′ predicted for the (n + 1) th image data.
Even when there are some differences between the actual optimum value of the quantization parameter and, in the range AR ₁ quantization parameter Q takes a large value, the difference between the corresponding data amount is small, the predicted quantization Even if the data amount obtained by quantizing and compressing the (n + 1) -th image data using the parameter Q _{n + 1} ′ exceeds the optimum data amount _Dobj , the data amount does not reach the limit data amount _Dmax and overflow occurs. Absent. That is, in the range where the quantization parameter Q is large, even if the amount of compressed data increases, the increase is limited by the limit data amount _Dmax and the target data amount _Dmax.
It can be absorbed by the difference (margin) from _obj , so that overflow does not occur.

On the other hand, in the case of a scene with a small amount of information,
Quantization parameter Q takes a small value, in the range AR ₂ quantization parameter Q takes a small value, as can be seen from Figure 2, the slope of the Q-D curve becomes steep. Therefore, even if the scene changes continuously, ((n +
1) Even though the image data changes slightly from the n-th image data), the actual optimal quantization parameter changes from time to time, and therefore, the quantization parameters Q _n , Predicted from data amount D _n
Quantization parameter Q for (n + 1) th image data
_{n + 1} 'may deviate considerably from the actual optimum value. When the predicted quantization parameter Q _{n + 1} ′ is shifted to a value larger than the actual optimum value, the compression rate increases and the data amount decreases, so that no overflow occurs. However, the predicted quantization parameter Q
_{If n + 1} ′ deviates considerably from the actual optimum value, the compression ratio becomes small, the data amount becomes considerably large, and the difference between the limit data amount D _max and the target data amount D _obj becomes larger.
(Margin) cannot be absorbed completely, and overflow exceeding the limit data amount _Dmax easily occurs.

[0029] Thus, in the range AR ₂ quantization parameter Q takes a small value, the fluctuation range of the data amount due to the change of the quantization parameter Q is large, the limit data amount D
without being completely absorbed by the difference (margin) between the _max and the target data amount D _obj, than an overflow occurs frequently, in the range AR _2, it is not preferable to vary each time the value of the quantization parameter Q.

Therefore, in this embodiment, in a range where the value of the quantization parameter Q does not become too small, that is, in a case where the scene is continuous, the fluctuation range of the data amount is limited to the limit (maximum) data amount D _max and the target data amount. Difference from data amount _Dobj
A limit value (minimum value) Q _min of the quantization parameter Q is set in a range that can be absorbed by (margin). In the example of FIG. 2, the Q _min,
Set to one value within the range of 16 ≦ Q ≦ 20. When the predicted quantization parameter Q _{n + 1} 'is less than Q _min is determined at a constant value Q _min the quantization parameter Q _{n + 1} for the (n + 1) th image data.

As a result, it is possible to effectively prevent an overflow from occurring in a small range of the quantization parameter Q. In particular, when the scene continuously changes, the reproduced moving image may temporarily stop or be disturbed. Can be prevented.

FIG. 3 is a flowchart for explaining the processing of the data compression apparatus of FIG. Referring to FIG.
Image compression unit 2, n-th reading the quantization parameter Q _n determined by the compression rate determination unit 16 to the image data of the (step S1), the performs quantization using the quantization parameter Q _n, the data amount Is compressed (step S2). When the data amount D _n of the compressed result is detected by the data amount detecting section 4 (step S3), and the compression ratio control unit 5, the (n + 1) -th quantization for the image data of the parameter Q _{n + 1} ', quantization parameter Q _n and the data amount D _n and the target data amount D _obj
(Step S4). Now, when the QD curve _Pn for the n-th image data is, for example, as shown in FIG. 2, Qn _{+ 1} ′ is obtained from Equation 4 as the QD curve _Pn for the _nth image data. It is predicted as a quantization parameter where the target data amount _Dobj intersects.

[0033] At this time, the quantization parameter Q is smaller range AR _2, even when the scene change is continuously
(QD curve P for (n + 1) th image data
_{(Even if n + 1} does not change much from the QD curve P _{n for the nth} image data), the optimal quantization parameter for the (n + 1) th image data is It varies considerably from the quantization parameter Q _n. Therefore, the range AR ₂ where the quantization parameter Q is small is
Then, it is difficult to predict the optimal quantization parameter for the (n + 1) th image data itself, and when the predicted quantization parameter Q _{n + 1} ′ becomes smaller than the optimal value, quantization is performed using this. When this operation is performed, the data amount exceeds the limit data amount _Dmax and an overflow occurs.

Therefore, the compression ratio determination unit 16 compares the predicted quantization parameter Q _{n + 1} ′ with the limit value Q _min (step S5), and determines whether the predicted quantization parameter Q _{n + 1} ′ is Q
If it is larger than _min , Q _{n + 1} 'itself is determined as the quantization parameter Q _{n + 1} for the (n + 1) -th image data (step S6), but the predicted quantization parameter Q _{n + 1} ' when less than Q _min determines the quantization parameter Q _{n + 1} for the (n + 1) th image data as Q _min (step S7).

After determining the quantization parameter Q _{n + 1} for the (n + 1) -th image data in this way, the process returns to step S1, and the image compression section 2 determines in step S6 in step S6 or S7. The obtained quantization parameter Q _{n + 1} is read, and in step S2, quantization is performed on the (n + 1) -th image data using the quantization parameter Q _{n + 1} .

As described above, in this embodiment, the quantization parameter Q _{n + 1} for the (n + 1) -th image data is predicted and determined by only one feedback control. The time required for the prediction can be shortened as compared with the related art, and the compression processing can be performed in real time. At this time, the predicted quantization parameter Q for the (n + 1) th image data is
_{When n + 1} ′ is smaller than the limit value _Qmin , the quantization parameter Qn _{+ 1} for the (n + 1) th image data is set to _Qmin.
By using this, the data can be compressed to the limit data amount _Dmax or less. That is, it is not necessary to cause an overflow, and for example, as described above, it is possible to reliably prevent a reproduced image from being stopped or disturbed in a continuous scene.

In this case, one frame (field) or two frames (field) is performed by performing quantization using a quantization parameter Q _min larger than the predicted quantization parameter Q _{n + 1} '. During this time, the image quality (resolution) of the reproduced image is degraded.However, compared to the case where the reproduced image is frozen or disturbed for a moment, even if the resolution is slightly Has no effect.

As described above, in the data compression apparatus of the present embodiment, even if the compression ratio (quantization parameter) becomes small,
It is possible to perform the compression processing in real time without causing the overflow and suppressing the deterioration of the image data quality.

In the above embodiment, the n-th frame (field) and the (n + 1) -th frame (field)
Although the DCT conversion is performed on the image data of one frame (field), the DCT conversion on the image data of one frame (field) is actually performed by converting the image data (digital image data) of one frame (field) into a plurality of pixels. (For example, a block of 8 × 8 pixels), a DCT is performed for each block, and a transform coefficient obtained for each block is quantized by a quantization parameter and compressed. However, the division for each block is not directly related to the present invention, and therefore, in the above description, for convenience, DCT transform is performed on image data of one frame (field), and the obtained transform coefficient is quantized. Quantization and compression are performed according to the quantization parameter.

In the above embodiment, the image data is
Although the data compression (quantization) is performed after performing the CT transformation, the present invention can be similarly applied to a case where an arbitrary transformation other than the DCT transformation is used.

In the above-described embodiment, the image compression unit 2 actually performs the compression processing using the quantization parameter Q by using the quantization matrix (i, j) as shown in Expression 1. At this time, when a quantization parameter is given, it takes a very long time to calculate the quantization matrix (i, j) by Equation 1 each time, which causes a problem in real-time processing. For the quantization parameter Q, a quantization matrix (i, j) is determined in advance, and these are stored in, for example, a ROM so as to correspond to the value of the quantization parameter Q. When the value of the quantization parameter Q is given, Is preferably read out from the ROM. In the above embodiment, the JPE
A quantization matrix (quantization parameter) defined by G
The compression is performed according to the following equation. However, as long as the data is compressed at a predetermined compression rate, not limited to the quantization matrix (quantization parameter) specified by JPEG, an arbitrary one can be used.

In the above-described embodiment, compression of image data which fluctuates with time, such as a moving image, will be described. The present invention is particularly effective in compressing image data. The present invention can be applied to compression of arbitrary data that fluctuates. Therefore, in the above embodiment,
Although it is assumed that data is compressed in units of one frame (one field), the present invention is applicable not only to one frame (one field) but also to a case where compression is performed in arbitrary time units.

[0043]

As described above, according to the present invention, the quantization parameter for the data at the next time is calculated based on the compression result obtained by performing the compression processing on the data at a certain time. a data compression apparatus for compressing data by a feedback control to predict when the next
The quantization parameter for the data at the point
The quantization parameter for the data at
Data compression and data overflow.
Target data volume set to be smaller than the shear limit data volume
And the amount predicted by feedback control
When Coca parameter is smaller than a predetermined quantization parameter, the quantization path for the data at the next time
The parameter is determined to the predetermined quantization parameter . In this case, the predetermined quantization parameter
Compress parameters with predicted quantization parameters
Of data amount from target data amount when
Is absorbed by the difference between the limit data amount and the target data amount.
Since set to a range that can, even when the quantization parameter is small, without causing an overflow, is converged to a large amount of data as possible the amount of data, performs compression processing by suppressing the degradation of data quality and in real time be able to.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of an embodiment of a data compression device according to the present invention.

FIG. 2 is a diagram for explaining an outline of processing of the data compression device of FIG. 1;

FIG. 3 is a flowchart showing a processing flow of the data compression device of FIG. 1;

FIG. 4 is a diagram showing QD curves for various images.

FIG. 5 is a diagram illustrating an example of a recording type of a recording apparatus.

FIG. 6 is a diagram showing a QD curve for two image data when a scene changes continuously.

[Explanation of symbols]

 2 Image compression unit 4 Data amount detection unit 5 Compression ratio control unit 16 Compression ratio determination unit 22 Buffer memory 23 Recording device

Continuation of the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H03M 7/30 H04N 5/92

Claims (1)

(57) [Claims] A data compression apparatus for compressing data by feedback control for predicting a quantization parameter for data at a next time based on a compression result obtained by performing compression processing on data at a certain time. Then , the quantization parameter for the data at the next point in time is
A quantization parameter for the data at some point, and
The amount of data resulting from compressing the data at the point in time
Eyes set to be smaller than the limit data amount at which flow occurs
And the quantization parameter predicted by the feedback control is a predetermined quantization parameter.
When smaller than data, is adapted to determine a quantization parameter for the data at the next point in the predetermined quantization parameter <br/>, in this case, the predetermined
The quantization parameter
Therefore, the target data amount of the data amount when compression is performed
Fluctuation range between the limit data amount and the target data amount
A data compression apparatus characterized in that the data compression apparatus is set in a range that can be absorbed by a difference .