JPH05153569A - Picture coder - Google Patents

Abstract

PURPOSE:To control a code quantity through the provision of a virtual buffer used to change a quantity step width to obtain the code quantity allocated to each picture in order to keep the picture quality for in-frame coding only and the picture quality for the selection of in-frame or inter-frame coding constant. CONSTITUTION:Any of I, P, B pictures is selected by a switch 29 in a code quantity control section 23 from a code quantity quantized by a quantizer 16 while being divided in the unit of macro blocks and outputted by a coder 17 and the selected picture is inputted to an allocation code quantity calculation section 24. The allocation code quantity calculation section 24 calculates the code quantity of each picture and calculates an allocation code quantity of each coded picture. The quantization step width is changed from the data so as to obtain a code quantity allocated to each picture and the resulting data are stored in virtual buffer sections 25, 26, 27 for I, P, B picture controlled independently for each picture.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding apparatus for highly efficiently coding and compressing image data, and more particularly to an image coding apparatus for controlling a code amount for recording the compressed data on a storage medium. ..

[0002]

2. Description of the Related Art Conventionally, a moving picture is referred to as an I picture (I-picture) for intraframe coding only, and a P picture for selecting intraframe coding and interframe coding in macroblock units. And B picture (P-picture
And B-picture), and by randomly arranging these images as shown in FIG. 8A, random access required for recording on the storage medium, A moving image coding method that achieves high-efficiency coding compression while enabling reverse playback has been proposed as an international standardization proposal.

The P picture and B picture mean a forward predictive coded image and a bidirectional predictive coded image, respectively. Further, the macroblock is a group of a plurality of blocks (8 × 8 pixels) in which a plurality of pixels are grouped together.

The above-mentioned international standardization proposal is for recording compressed encoded data in a storage medium having a constant transfer rate, and overflows or underflows of a buffer memory for storing the encoded data in order to obtain a constant transfer rate. The condition for no flow to occur, that is, the upper limit and the lower limit of the generated code amount per frame are defined.

Several methods of controlling the amount of generated codes per frame have been proposed. For example, JP-A-2-305
In No. 083, the allocated code amount for the intra-frame coded image and the inter-frame coded image is determined so that the image quality becomes uniform based on the amount of data remaining in the transmission buffer memory, and becomes this assigned code amount. Thus, the quantization scale is controlled based on the data remaining in the transmission buffer memory.

Quantization scale STEP = [(M _R / M _T ) × α] + β where M _T is the transmission buffer memory capacity, M _R is the transmission buffer memory remaining amount, and α is a step that the quantization scale STEP can take. The numbers and β represent adjustment values.

In addition, the simulation model proposed in the international standardization proposal is controlled as follows. That is, the allocated code amount of each picture is determined from the ratio of the newest generated code amount of each picture (I, P, B picture). Then, as shown in FIG. 9A, the amount of code generated in units of a region where several macroblocks called slices are gathered in the line direction in a pixel of one frame, and is given to the region in advance. An error from the assigned code amount is accumulated in the virtual buffer. The quantization scale, which is proportional to the occupancy of this virtual buffer,
The quantization scale of the slice to be coded next is used. As a result, the amount of generated codes in one frame is controlled.

It should be noted that there is only one virtual buffer, as shown in FIG.
As shown in (b), the quantization scale used for the first slice of the P picture to be encoded next is the quantum scale corresponding to the virtual buffer occupancy immediately after the end of encoding of the last slice of the I picture. A digitization scale is used. Here, the coding of each picture is performed in the order as shown in FIG.

[0009]

As described above, when one transmission buffer and one virtual buffer are used to control the code amount, the picture quality of each picture is made uniform.

Further, in the above international standardization proposal, the coding noise generated in the I picture is caused by the dependency relationship of the P picture predicted from the decoded data of the I picture and the B picture predicted from the I or P picture. Affects the P picture, and the coding distortion generated in the I or P picture affects the B picture, and a visually noticeable deterioration remains in each picture.

In particular, when the picture quality of each picture is made uniform within the determined transmission rate (that is, the average quantization scale of each picture is the same), the picture quality of the I picture is as follows. That is, when the picture quality of I pictures is weighted for each picture (the average quantization scale of each picture is not the same), specifically, the picture quality is lowered in the order of I picture, P picture, and B picture. Coding distortion increases as compared with the case of going forward. In a P picture in which prediction is performed from a decoded image of an I picture having a large amount of coding distortion, this coding distortion is dragged. This is B
The same applies to pictures.

Further, in the convergence stage until the picture quality of each picture becomes uniform, the fluctuation of the quantization scale in each picture becomes large. This means that there is a large variation in the image quality within the picture, and the phenomenon that the image quality of the visually insignificant part is improved and the image quality of the important part is degraded occurs, so that the coding efficiency is improved. It's not good.

The present invention has been made in view of the above-mentioned problems, and it is possible to realize I picture without making the picture quality of all pictures uniform.
Image coding capable of maintaining the weighted state of the picture quality of each of the picture, P picture, and B picture, reducing the fluctuation of the picture quality within each picture, and improving the picture quality of a visually important portion within each picture. The purpose is to provide a device.

[0014]

That is, according to the present invention, there is provided a first image for performing only intraframe encoding of a digital image signal and a second image for selecting intraframe and interframe encoding. In the image coding apparatus for assigning a code amount to each of the first and second images so as to have a constant code amount and performing high-efficiency coding compression, the code assigned to each image It is characterized in that a plurality of virtual buffer units for changing the quantization step width so as to become a quantity are provided, and the plurality of virtual buffer units are independently controlled for each of the images.

[0015]

According to the present invention, the digital image signal is composed of an image for which only intra-frame coding is performed and an image for which intra-frame and inter-frame coding is selected for coding. In an apparatus that performs high-efficiency coding and compression so that the image quality becomes constant, the image quality of an image that is only intra-frame coded and the image quality of an image that is coded by selecting intra-frame or inter-frame coding. In order to keep each constant,
A virtual buffer unit used for changing the quantization step width so as to have the code amount assigned to each image is separately provided for each image and independently controlled. As a result, the image quality of each image can be made uniform separately, and the fluctuation of the quantization step width in the image can be suppressed.

Further, the code for each image or the virtual buffer occupancy at the end time is compared with the average virtual buffer occupancy,
The initial quantized step size is calculated using the larger occupancy next time, and only when the average virtual buffer occupancy is selected, the virtual occupancy at the end of encoding is calculated from the average virtual buffer occupancy. The difference amount obtained by subtracting the buffer occupancy amount is added to the assigned code amount of the next encoded image. In particular, the difference amount is distributed mainly to the visually important portion in the image.

[0017]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of an image coding apparatus according to the present invention, and FIG. 2 is a diagram showing an example in which digitized image data is divided into macroblocks. It should be noted that, in the following, I
A case where the picture, P picture, and B picture are configured will be described.

In FIG. 1, the image data divided into macroblock units includes a subtractor 11, a switch 12 and a predictor.
Entered in 13. The predictor 13 outputs the subtractor 11 and the switch 14. The output of the switch 12 is DC
After passing through T15, the quantizer 16 quantizes the transform coefficient and sends it to the transmission buffer 18 via the encoder 17 for storage. The output from the quantizer 16 is the inverse quantizer 19, inverse D
It is sent to the predictor 13 via the CT 20 and the adder 21. 22
Is a prediction mode determiner.

The code amount control section 23 receives the code amount generated from the encoder 17 and the specified transmission rate value. The code amount control unit 23 includes an assigned code amount calculation unit 24, an I picture virtual buffer unit 25, a P picture virtual buffer unit 26, and a B picture virtual buffer unit.
27, a quantization scale calculator 28, and switches 29 and 30. The details of the code amount control unit 23 will be described later. Next, the operation of the embodiment will be described.

As shown in FIG. 2, the image data divided into macroblock units is input to the subtractor 11, the switch 12 or the predictor 13. The macroblock data input to the predictor 13 and the motion-compensated prediction macroblock are output to the subtractor 11. In the subtractor 11,
The prediction macroblock data is subtracted from the input macroblock data to generate prediction error macroblock data. Then, this generated data is sent to the switch 12
Is output to.

In the switch 12, the prediction error macroblock data and the input macroblock data are selected by the judgment signal from the prediction mode judging device 22, and the result is output to the DCT 15. Here, when the encoded image is an I picture, the input macroblock data is forcibly selected.

The macroblock data input to the DCT 15 is divided into "1 to 6" blocks shown in FIG. 2, and the discrete cosine transform is performed for each block (8 × 8 pixels). Transform coefficients are generated. This transform coefficient is linearly quantized by the quantizer 16. The quantization step size used for each transform coefficient is different. Generally, the step width corresponding to the low frequency component is small, and the step width is increased as it goes to the high frequency component.

Specifically, as shown in FIG. 3, the quantization step width weight q ₀₀ corresponding to each transform coefficient is arranged in the same quantization matrix as the transform coefficient in the quantization matrix Q. By multiplying _S , the quantization step size of each transform coefficient is determined.

Here, the quantization matrix is the switch 12
The two preset matrices are switched according to the data (input macroblock and prediction error macroblock) selected by. The quantization scale Q _S is used to control the code amount, and the value of Q _S is changed in units of slices shown in FIG.

As shown in FIG. 4, the quantized transform coefficient undergoes zigzag scanning in order from the low frequency component and is output to the encoder 17 as one-dimensional data. In the encoder 17, a variable length code is assigned as a pair with a continuous number having a quantized value of "0" and a value immediately after that, which is not "0", to perform encoding. The data encoded here is accumulated in the transmission buffer 18.

The quantized transform coefficient from the quantizer 16 is inversely quantized by the inverse quantizer 19, and this data is inverse discrete cosine transformed by the inverse DCT 20. The switch 14 determines whether or not the data from the inverse DCT 20 and the motion-compensated prediction macroblock which is the output data from the predictor 13 are selected by the determination signal output by the prediction mode determiner 22. ..

The predictive macroblock or data without input is added by the adder 21 to generate a decoded image. This decoded image is stored in a frame memory (not shown) in the predictor 13 only in the case of I picture and P picture. This stored data serves as a reference for the image to be motion-compensated and predicted next.

As described above, the code amount controller 23 receives the code amount generated from the encoder 17 and the specified transmission rate value. And
The output of the code amount control unit 23 is a quantization scale Q _S , and this value Q _S is output to the quantizer 16 to change the quantization step width. Here, the operation of the code amount control unit 23 will be specifically described.

The code amount output from the encoder 17 is
Any one of an I picture, a P picture and a B picture is selected by the switch 29 and input to the allocation code amount calculation unit 24. The allocated code amount calculation unit 24 calculates the code amount of each picture, and calculates the allocated code amount of each picture to be encoded next from these values and the transmission rate. These data are stored in the I-picture virtual buffer unit 2
5, input to the P picture virtual buffer unit 26 and the B picture virtual buffer unit 27. Each virtual buffer unit 25, 26,
In 27, the assigned code amount Ac is distributed in slice units. The code amount output from the encoder 17 is selected by the switch 29 for each of I picture, P picture, and B picture.

Virtual buffer units 25, 26, 27 for each picture
Then, the generated code amount Gc for the slice is obtained, and an error from the assigned code amount Ac, Ec = Gc-Ac, is accumulated in the virtual buffer unit for each picture. The occupied amount of the virtual buffer unit immediately after the storage is output to the switch 30 from each of the picture virtual buffer units. Then, the data of the currently encoded picture is selected by the switch 30 from these data, and is output to the quantization scale calculation unit 28.

In the quantization scale calculator 28, the quantization scale corresponding to the virtual buffer occupancy input based on the proportional relationship between the virtual buffer occupancy and the quantization scale as shown in FIG. It is output to the digitizer 16. In the quantizer 16, the quantization step width of each transform coefficient is determined by the quantization scale and the quantization matrix,
Quantization is performed.

FIGS. 6A to 6C show examples of the virtual buffer section of each picture. The initial occupancy of the virtual buffer unit of each picture is set in advance, or the initial occupancy of only the I picture is set, and the initial occupancy of the P picture and the B picture are respectively coded one time earlier. The occupancy of the virtual buffer unit at the time when the last slice of the encoded picture is encoded may be used. That is, it operates as a virtual buffer unit for each picture.
For example, the occupancy of the I-picture virtual buffer unit 25 is used as the initial occupancy of the P picture, and the occupancy of the P-picture virtual buffer unit 26 is used as the initial occupancy of the B picture.

As a result, the generated code amount is controlled so as to be the allocated code amount for each picture, and the quantization scale is not affected by the difference in the generated code amount between each picture. Fluctuations in are small. Also, the average quantization scale for each picture can be held substantially constant (assuming that there is no scene change). As a result, the image quality of the I picture is always improved, and the P picture is reduced by reducing the coding distortion.
The influence of the coding distortion of the I picture on the B picture can be reduced, and the image quality can be improved.

Further, in each picture virtual buffer section,
The average value of the virtual buffer occupancy amount for each slice of the picture of the same type encoded immediately before is obtained. This average virtual buffer occupation amount Ao is compared with the virtual buffer occupation amount Bo immediately after encoding the last slice. Where Ao>
In the case of Bo, the quantization scale of the first slice of each picture to be coded next is calculated using Ao. On the other hand, when Ao ≤ Bo, the quantization scale is calculated using Bo.

When Ao> Bo, Ao is used to calculate the quantization scale, and at the same time, the occupied amount of the virtual buffer section is changed from Bo to Ao. The remainder Ao-Bo of the occupied amount at this time is distributed to the allocated code amount of the slice (for example, the central portion of the screen) of the important portion of each picture to be encoded next.

As a result, the defect that the quantization scale becomes large in the central portion of the picture generated when Ao> Bo, that is, the image quality deterioration in the central portion can be reduced. On the contrary, when Ao ≤ Bo, the quantization scale in the central portion is not large in the picture,
In this case, the above-mentioned drawbacks do not occur.

This process is specifically shown in FIG. FIG. 7 is a block diagram showing a main part of the code amount control unit 23 of FIG. The code amount Gc output from the slice generation code amount calculation unit 31 and the code amount Ac output from the slice allocation code amount calculation unit 32 are subtracted by the subtractor 33. The allocation error value of this subtraction (Gc-Ac) is the virtual buffer
It is added to the occupancy of 34 to make a new virtual buffer occupancy. This virtual buffer occupancy is input to the slice average virtual buffer occupancy calculation unit 35, and the average virtual buffer occupancy within one picture is calculated.

The occupancy amount Bo of the virtual buffer 34 at the time when the encoding of one picture is completed and the average virtual buffer occupancy amount Ao are input to the selector 36. The selector 36 selects the larger one of the two values and outputs it to the quantization scale calculator 28 (see FIG. 1). When Ao> Bo, the difference amount of Ao-Bo is output to the slice allocation code amount calculation unit 32. In the slice allocation code amount calculation unit 32, the difference amount of Ao-Bo is distributed to the important part (for example, the central part of the image) of the same type of picture to be encoded next. By performing the above processing, it is possible to prevent image quality deterioration in the central portion of the image.

[0040]

As described above, according to the present invention, the generated code amount is controlled so that the ratio of the image quality of the intra-frame coded image and the inter-frame coded image becomes constant, and the variation of the image quality within each image. To reduce I picture, P picture, B picture without making the picture quality of all pictures uniform.
(EN) Provided is an image encoding device capable of maintaining the weighted state of each image quality of a picture, reducing the fluctuation of the image quality within each picture, and improving the image quality of a visually important portion within each picture. it can.

[Brief description of drawings]

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

FIG. 2 is a diagram showing an example in which digitized image data is divided into macroblocks.

FIG. 3 is a diagram illustrating determination of a quantization step width of each transform coefficient.

FIG. 4 is a diagram showing a state in which a quantized transform coefficient is sequentially subjected to zigzag scanning from a low frequency component to a high frequency component.

FIG. 5 is a diagram showing a proportional relationship between a virtual buffer occupation amount and a quantization scale.

6A to 6C are diagrams showing examples of virtual buffer units for I-pictures, P-pictures, and B-pictures, respectively.

7 is a block diagram showing a main part of a code amount control unit 23 of FIG.

8A and 8B are I-pictures and P-pictures in the related art.
It is the figure which showed the example which arranged the picture and the B picture periodically, (b) is the figure which showed the encoding order of the picture arranged like (a).

9A and 9B are diagrams showing an example of accumulating each picture in a virtual buffer according to a conventional technique.

[Explanation of symbols]

11 ... Subtractor, 12, 14, 29, 30 ... Switch, 13 ... Predictor,
15 ... DCT, 16 ... Quantizer, 17 ... Encoder, 18 ... Transmission buffer, 19 ... Inverse quantizer, 20 ... Inverse DCT, 21 ... Adder,
22 ... Prediction mode determiner, 23 ... Code amount control unit, 24 ... Allocated code charge calculation unit, 25 ... Virtual buffer for I-picture, 26 ...
P picture virtual buffer unit, 27 ... B picture virtual buffer unit, 28 ... Quantization scale calculation unit.

Claims (4)

[Claims] 1. A first image for performing only intra-frame encoding of a digital image signal, and a second image for performing intra-frame and inter-frame encoding by selecting encoding, and having a constant code amount. In the image coding apparatus for allocating a code amount to each of the first and second images so as to perform high-efficiency coding compression, the quantization step width is set so that the code amount is allocated to each image. An image coding apparatus, comprising: a plurality of virtual buffer units for changing the above, wherein the plurality of virtual buffer units are independently controlled for each of the images. 2. The control of the quantization step width of the virtual buffer unit is performed for each image by accumulating an error between a generated code amount and an assigned code amount for each of the first image and the second image. The image encoding device according to claim 1, wherein the image encoding device is performed separately according to its occupation amount. 3. The virtual buffer unit compares the virtual buffer occupancy at the end of encoding for each image with the average virtual buffer occupancy for each image, and determines the larger occupancy. The initial quantization step size of the next encoded image is calculated by using it, and only when the average virtual buffer occupancy is selected, the virtual buffer occupancy at the end of encoding is subtracted from the average virtual buffer occupancy. The image coding apparatus according to claim 2, wherein the difference amount is added to the assigned code amount of the next coded image. 4. The image coding apparatus according to claim 3, wherein the virtual buffer unit adds the allocated code amount of the difference amount by predominantly distributing it to a visually important portion in the image.