WO1997042767A1

WO1997042767A1 - Process for quantization and a process for inverse quantization of transform coding coefficients of a video data stream

Info

Publication number: WO1997042767A1
Application number: PCT/DE1997/000859
Authority: WO
Inventors: Claus Schneider
Original assignee: Siemens Aktiengesellschaft
Priority date: 1996-05-06
Filing date: 1997-04-28
Publication date: 1997-11-13
Also published as: DE19618117C1

Abstract

This invention concerns a process for quantization and a process for inverse quantization which are used for image encoding processes. This involves determination (301) of a first intermediate matrix (Y1) by multiplying a quantization matrix by at least one scalar factor. The first intermediate matrix (Y1) is temporarily stored (302) and is multiplied by a first data matrix, the elements of which are given by transform coding coefficients.

Description

description

Method for quantization and method for inverse quantization of transformation coding coefficients of a video data stream

In the context of telecommunications, which are becoming increasingly important these days, especially in the field of multimedia, the coding and compression of data to be transmitted, for example image data, is becoming increasingly important. The data should be encoded in such a way that the greatest possible compression of the information is achieved with as little loss of information as possible.

Various methods for coding a video data stream are known, for example block-based image coding methods such as MPEG (MPEG 1 or MPEG 2) [1], the H.261 standard [2] or JPEG [3]. Various so-called object-based image coding methods are also known.

In the case of the image coding method, the transformation coding coefficients are usually quantized for transformation coding coefficients which result from a transformation coding which is applied to the image to be coded. Furthermore, an inverse quantization is used for the reconstruction of the video data stream on the quantized coefficients, the result values of which are subjected to an inverse transformation coding for the final reconstruction of the digital image.

An efficient and rapid quantization or inverse quantization is of considerable importance in order to ensure a high data rate of digital images to be processed. Usually, particularly in block-based image coding methods, the quantization is carried out in such a way that elements of a data matrix which has the transformation coding coefficients during quantization and coefficients quantized during inverse quantization are multiplied by a predeterminable scalar factor and are temporarily stored in an intermediate matrix. The elements of the intermediate matrix are multiplied by elements of a predetermined quantization matrix.

This procedure hides v. a. the disadvantage in itself that, as will be described in the following, although the elements of the quantization matrix and the scalar factor remain unchanged for several successive blocks, the multiplications are carried out anew for each image block. This leads to a considerable number of multiplications which are actually not necessary, which in turn leads to an increased power loss when the known method is carried out. Furthermore, an increased computing time is required than is the case with the method according to the invention, as will be described in the following.

The invention is therefore based on the problem of specifying a method for quantization and for inverse quantization of transformation coding coefficients, wherein less computing time is required to carry out the quantization or for inverse quantization than in the known method and thus the methods are carried out more cost-effectively can be led.

The problem is solved by the method according to claim 1 and by the method according to claim 2.

A first method is used in the method for quantizing transformation coding coefficients of a video data stream

Intermediate matrix determined, which results from the multiplication of the quantization matrix by at least one scalar Factor. The first intermediate matrix is temporarily stored and multiplied by a data matrix which contains the transformation coding coefficients.

Since both the at least one scalar factor and the elements of the quantization matrix remain unchanged over several successive blocks or macro blocks to be quantized or inversely quantized, this ensures that a smaller number of operations is required, than is possible with the known arrangement. This also leads to a saving of the power loss required, which likewise leads to a considerable cost saving.

The same advantages also apply to the method for inverse quantization, in which a second intermediate matrix is determined, which results from the multiplication of a quantization matrix by at least one scalar factor. The second intermediate matrix is also temporarily stored and multiplied by a second data matrix that contains quantized coefficients.

Advantageous developments of the invention result from the dependent claims.

To further accelerate the implementation of the methods according to the invention and to reduce the memory requirement of the arrangements, it is advantageous to determine and / or temporarily store only predeterminable coefficients of the first intermediate matrix and / or the second intermediate matrix. This procedure both reduces the number of operations required, since a smaller number of elements of the intermediate matrices only have to be processed, and furthermore the space requirement is reduced, since only a smaller number of elements of the intermediate matrix are stored Need to become. In a development of the method, it is advantageous to reset a quantization result memory for storing result values of the quantization or the inverse quantization to the value 0 at the beginning of each quantization or inverse quantization and only to carry out multiplications of elements of the intermediate matrices, which have the value 0 and / or the value 1. This procedure results in a further reduction in the power loss required and a further reduction in the computing time required for quantization, since unnecessary multiplications with matrix elements which have the value 0 or the value 1 are avoided.

The method can easily be used in any block-based image coding method, for example in one of the methods for image coding mentioned above.

An exemplary embodiment of the invention is shown in the figures and is explained in more detail below.

Show it

1 is a block diagram showing an arrangement for image coding;

Fig. 2 is a block diagram in which a sketch of the

Arrangement for image decoding is shown;

3 shows a flow diagram in which the method steps for quantization are shown;

4 shows a flow chart in which the method steps of the method for inverse quantization are shown;

Fig. 5 is a flowchart in which the method for

Quantization with some further training of the driving are shown;

6 shows a flow chart in which the use of the method for quantization and / or for inverse quantization in the context of the image coding is shown;

7 shows a flow chart in which the use of the method according to the invention in the context of image decoding of a video data stream is shown;

Fig. 8 is a sketch showing two computers, images recorded by a camera are encoded in a first computer and transmitted to a second computer, where they are decoded.

1 shows a block diagram in which an arrangement for image coding is described.

The arrangement has at least the following components: a first means DCT for performing a transformation coding, for example a discrete cosine transformation; a quantization unit Q; an inverse quantization unit IQ; a second means IDCT for performing inverse transform coding; - an addition unit AE; a memory SP; a third means BS for carrying out a motion estimation; a subtraction unit SE.

Here, the first means DCT and the quantization unit Q form a forward path VP. The inverse quantization Unit IQ and the second means IDCT form a backward path RP.

Furthermore, a fourth means VLC is provided for performing channel coding of the quantized transformation coefficients, possibly with an additional unit for error detection and / or error correction of bit errors.

An output of the subtraction unit SE is coupled to an input of the first means DCT. An output of the first means DCT is coupled to an input of the quantization unit Q. An output of the quantization unit Q is coupled to an input of the inverse quantization unit IQ. Furthermore, the output of the quantization unit Q is coupled to an input of the fourth means VLC. An output of the inverse quantization unit IQ is coupled to an input of the second means IDCT. An output of the second means IDCT is coupled to a first input of the addition unit AE. An output of the addition unit AE is coupled to an input of the memory SP. An output of the memory SP is coupled to a first input of a third means BS.

A first output of the third means BS is coupled to a second input of the memory SP. With this coupling, memory addresses ADR are transferred from the third means BS to the memory SP. The memory addresses ADR indicate the memory addresses ADR required by the third means BS to carry out the motion estimation.

Furthermore, a second output of the memory SP is coupled to a second input of the subtraction unit SE and to a second input of the addition unit AE. A video data stream VD to be coded, which in the case of block-based image coding methods has individual image blocks, is applied to the first input of the subtraction unit SE. The video data stream VD to be encoded is also applied to the third means BS, where the video data stream VD is used in the context of a motion estimation.

The configuration of the quantization unit Q and / or the inverse quantization unit Q is explained in detail in the course of the description of the methods. The design of the quantization unit Q or the inverse quantization unit IQ is such that the method for quantization described below or the method for inverse quantization can be carried out.

2 shows an arrangement for image decoding. The arrangement for image decoding has at least the following components:

A fifth means VLD for performing an inverse channel coding of quantized transformation coding coefficients; the inverse quantization unit IQ; - The second means IDCT for performing the inverse transformation coding; the addition unit AE; the memory SP; the third means BS for motion estimation.

A video data stream ZVD to be reconstructed, which has quantized, channel-coded transformation coding coefficients, is fed to the fifth means VLD for inverse channel coding.

After the inverse channel coding VLD has been carried out, the quantized transformation coding coefficients are fed to the inverse quantization unit IQ, which is coupled to the fifth means VLD. After the inverse quantization IQ, the inverse quantized transformation coding coefficients are fed to the second means IDCT, the input of which with the output of the inverse quantization unit IQ is coupled. An output of the second means IDCT is coupled to a first input of the addition unit AE. An output of the memory SP is coupled to a second input AE. A temporally preceding, reconstructed image of the reconstructed video data stream RVD is applied to the second input of the addition unit AE. An output of the addition unit AE is coupled to an input of the memory SP. Furthermore, an output of the memory SP is coupled to an input of the third means BS. A second input of the third means BS for motion estimation is coupled to an output of the memory SP, via which the reconstructed images which have preceded the time and are stored in the memory SP are fed in to carry out a motion estimation in the third means BS.

The configuration of the inverse quantization unit IQ is shown in detail in connection with the method according to the invention for inverse quantization.

In a first step 300 of the method for quantization before each quantization, it is checked whether at least one scalar factor, which is multiplied by a quantization matrix, or the quantization matrix has changed.

If the scalar factor and / or the quantization matrix has changed, a first intermediate matrix Y1 is determined 301.

This is done by multiplying the quantization matrix, the structure of which, for example, from documents [1], [2], [3],

[4] is known, with at least one scalar factor (cf. FIG. 3).

The first intermediate matrix Y1 is temporarily stored 302. The first intermediate matrix Yl remains unchanged until either the at least one scalar factor or the quantization matrix changes. It has been found that the quantization matrix and the at least one scalar factor for an average of 10 to 20 successive image blocks can be kept unchanged.

If the check in the first method step 300 for quantization shows that both the at least one scalar factor and the quantization matrix have remained unchanged, the first intermediate matrix Y1 determined for the last quantization used is used directly for the method step described below.

In a third step 303, the first intermediate matrix Y1 is multiplied by a first data matrix which contains the transformation coding coefficients. The transformation coding coefficients are formed by a transformation coding which is applied to the video data stream to be coded.

4 shows a flowchart which describes the individual method steps of the method for inverse quantization.

In the inverse quantization of quantized transformation coding coefficients of the video data stream ZVD to be reconstructed, in a first step 400 it is checked before each quantization whether at least one scalar factor, which is multiplied by a quantization matrix, or the quantization matrix has changed.

If the scalar factor and / or the quantization matrix has changed, a second intermediate matrix Y2 is determined 401. The second intermediate matrix results in the same way as the first intermediate matrix Y1, that is to say from the multiplication of the quantization matrix by at least one scalar factor.

In a second step 402, the second intermediate matrix Y2 is temporarily stored. The statements made above The duration during which the quantization matrix and the scalar factor remain unchanged on average also apply to this method.

In a last step 403, the second intermediate matrix Y2 is multiplied by a second data matrix. The elements of the second data matrix are coefficients which are present after the inverse channel coding, that is to say quantized transformation coding coefficients which are referred to below as quantized coefficients.

FIG. 5 shows a further development of the method for quantization shown in FIG. 3, in which at the beginning of the quantization a first quantization result memory for storing result values of the quantization is reset to the value 0 501.

Since the first quantization result memory has been reset, only operations for elements of the first data matrix that have a non-zero value have to be carried out in this development.

After the storage 302 of the first intermediate matrix Y1, in a further embodiment of the method it is checked for each element of the first data matrix whether the element has the value 1 502.

If this is not the case, the corresponding multiplication of the elements of the first intermediate matrix Y1 is carried out 303 with the elements of the first data matrix.

If this is the case, however, the value of the corresponding element of the first intermediate matrix Y1 will simply be transferred to a result matrix. In this case, further multiplication operations do not have to be carried out, which leads to a considerable saving in computing time. The values of the quantized transformation coding coefficients are stored in the result matrix.

Although the development in FIG. 5 relates only to the method of quantization, the corresponding steps can also be used accordingly for inverse quantization without any fundamental changes.

Furthermore, in a further development of the method for quantization and the method for inverse quantization, it is provided that only predeterminable coefficients of the first intermediate matrix Y1 and / or the second intermediate matrix Y2 are determined and / or buffered.

This is advantageous since it has been found that only on average approximately 10 to 15% of all elements of the first data matrix and / or the second data matrix Y2 have a value that is not equal to zero. The fact that only a part of all elements of the first intermediate matrix Y1 or the second intermediate matrix Y2 is taken into account, the

Implementation of the procedures accelerated considerably. Furthermore, a possible reduction in the memory requirement is possible if a first buffer store for storing the elements of the first intermediate matrix and / or a second buffer store for storing the elements of the second intermediate matrix Y2 is configured as a cache memory. The cash hit rate can be increased further by refining the cache granularity to the elements of the first intermediate matrix Y1 or the second intermediate matrix Y2 for special, predeterminable elements of the intermediate matrices Y1, Y2. A corresponding strategy of the cache memory can considerably reduce the size of the first buffer store and the second buffer store required, as a result of which the costs required for implementation are considerably reduced.

6 shows the use of the methods in a method for image coding in a flow chart. There- at DCT, transformation coding coefficients are formed 601 in the first means. The transformation coding coefficients are quantized 602 in the quantization unit Q and the quantized transformation coefficients are subjected to an inverse quantization 603 in the inverse quantization unit IQ.

In the second means IDCT, an inverse transformation coding is carried out on the inversely quantized transformation coding coefficient, whereby a reconstructed video data stream RVD is formed 604. Furthermore, a motion estimation is performed 605 for the reconstructed video data stream RVD. In a last step the reconstructed video data stream RVD is subtracted from the video data stream VD, so that only the difference between the video data stream VD and the reconstructed video data stream RVD is fed 606 to the first means DCT for transform coding.

In Fig. 7 is the use of the inverse method

Quantization shown as part of image decoding. In this case, an inverse quantization of the quantized, inverse channel-coded transformation coding coefficients is carried out 701 in the inverse quantization unit. In the second means IDCT, an inverse transformation coding is carried out on the inversely quantized transformation coding coefficient and the reconstructed video data stream RVD is carried out Formed 702. In the third means BS, a motion estimation is carried out 703 for the reconstructed video data stream RVD.

Both the arrangement for image coding and the arrangement for image decoding can have further components which are not shown in FIGS. 1 and 2. These components are provided, for example, for the standard implementation of scan methods, inverse scan methods, vector quantization, run-length coding or else one Run-length decoding. These additional components and the resulting additional process steps, which are usually in the z. For example, standardized block-based image coding methods described above can be used without restrictions in the method according to the invention and in the arrangements according to the invention. Since these method steps are carried out in the standardized methods, but are not essential for the actual invention, they are not described further here.

The method was only shown in the context of a block-based image coding method in the exemplary embodiment, but can also be used in any object-based image coding method.

8 shows an arrangement in which the methods are used. In a first computer RI, for example realized by the arrangement for image coding, an image is encoded in the manner described above. The image is recorded, for example, by a camera K. The first computer RI is coupled to a second computer R2. The first computer RI transmits the coded picture to the second computer R2, where the picture is decoded. The decoded picture is now displayed, for example, on a screen BS2 of the second computer R2 to a second viewer B2. It is further provided that the coded image is not only transmitted, but also decoded in the first computer RI, which is necessary in any case in the block-based method for determining the difference information actually transmitted, and on a screen BS1 of the first computer RI ei ¬ a first viewer B1 is shown. The following publications were cited in this document:

[1] D. Le Gall, A Video Compression Standard for Multimedia Applications, Communications of the ACM, Vol. 34, No. 4, pp. 47-58, April 1991

[2] Ming Liou, Overview of the px64 kbit / s Video Coding Standard, Communications of the ACM, Vol. 34, No. 4, pp. 60-63, April 1991

[3] G. Wallace, The JPEG Still Picture Compression Standard, Communications of the ACM, vol. 34, no. 4, pp. 31-44, April 1991

Claims

claims

1. Method for quantizing transform coding coefficients of a video data stream,

a first intermediate matrix (Y1), which results from the multiplication of a quantization matrix and at least one scalar factor, is determined (301),

- in which the first intermediate matrix (Yl) is temporarily stored (302), and

- in which the first intermediate matrix (Y1) is multiplied by a first data matrix which contains the transformation coding coefficients (303).

2. Method for inverse quantization of quantized coefficients of a video data stream to be reconstructed,

- in which a second intermediate matrix (Y2), which results from the multiplication of a quantization matrix and at least one scalar factor, is determined (401), - in which the second intermediate matrix (Y2) is temporarily stored (402), and

- in which the second intermediate matrix (Y2) is multiplied by a second data matrix which contains the quantized coefficients (403).

3. The method according to claim 1 or 2, in which only predeterminable coefficients of the first intermediate matrix (Yl) and / or the second intermediate matrix (Y2) are determined and / or temporarily stored.

4. The method according to claim 1, wherein a first quantization result memory for storing result values of the quantization and / or a second quantization result memory for storing result values of the inverse quantization at the beginning of each

Quantization or inverse quantization is reset to zero and only multiplication of elements of the first data matrix or the second data matrix (Y2), which do not have the value zero and / or the value one, are carried out with elements of the first intermediate matrix (Yl) or the second intermediate matrix (Yl).

5. Use of the method according to one of claims 1 to 4 for image coding,

- in which transformation coding coefficients are formed in a first means (DCT), - in which the transformation coding coefficients are quantized (602) in a quantization unit (Q),

- in which the quantized transformation coding coefficients are subjected to inverse quantization (603) in an inverse quantization unit (IQ), - in which, in a second means (IDCT), an inverse transformation coding (IDCT) on the inverse quantized transformation coding coefficient a reconstructed video data stream (RVD ) is formed (604),

in which a movement estimation on the reconstructed video data stream (RVD) is carried out in a third means (BS) (605),

- In which the reconstructed video data stream (RVD) is subtracted from the video data stream, so that only the difference between the video data stream and the reconstructed video data stream (RVD) is fed to the transformation coding (606).

6. Use of the method according to one of claims 1 to 4 for image decoding, - in which in an inverse quantization unit (IQ) quantized transformation coding coefficients are subjected to an inverse quantization (602), wherein

a reconstructed video data stream (RVD) is formed in a second means (IDCT) of an inverse transformation coding (IDCT) on the inversely quantized transformation coding coefficient (604), in which a movement estimation on the reconstructed video data stream (RVD) is carried out in a third means (BS) (605),