DE19603808C1 - Block-based compression image decoder - Google Patents

Abstract

In the method proposed, the steps for decoding uncoded macroblocks (UMB) are carried out in parallel with the decoding of encoded macroblocks (MB). This enables considerable savings in time to be achieved during decoding.

Description

The invention relates to a method within the scope of the Bewe compensation in block-based coding methods is set. With these methods of image processing An image is divided into macro blocks that have a predefinable type number of image blocks. When coding, the Image areas of the macro blocks or image blocks usually an encoding transformation, for example a disc subjected to ten cosine transformation (DCT). The ge transformation coefficients formed are then in a quantization level and a scan procedure subjected, for example, to the so-called Zig-Zag scan Process or the so-called alternate scan process. Furthermore, a run length code is created after the scanning process tion (RLC) and then a variable length Coding (VLC).

An enumeration that is in no way definitive of block-based coding methods is:

• - MPEG 1, described in [1]
• - MPEG 2, described in [2],
• - H.261 standard and
• - JPEG, described in [3].

A pre-processing video data compression system processing level, an encoder and a post-processing level e.g. B. from document US 5 367 629 known.

Usually, when coding an image it becomes an image blocks provided that contain both luminance information (Luminance blocks) as well as chrominance information (Chrominance blocks) over the encoded image area.

The now coded coefficients have to be reconstructed of the respective coded picture to the previous coding steps are subjected to inverse procedural steps.  

This means the use of a variable length decoding (VLD) to the coded coefficients and one the so-called run length decoding (RLD).

After carrying out a scan method ( 15 ) which is inverse to the scanning method used for coding, a quantization inverse to the quantization used for coding and an inverse coding transformation (IDCT) is carried out.

In order to save the transmission capacity required Chen, the principle of Prediction and a so-called motion compensation and Be motion estimation applied. This is when coding of a picture does not always encode the entire picture information, but sometimes only the difference information.

Images whose image information is completely encoded are referred to as support pictures (I-pictures).

So-called prediction bil between these I-pictures the (P-pictures) coded, whereby only the difference information mation coded to the previous coded picture which is reconstructed in the coding unit.

So-called interpolation images (B- Images) that are bidirectional from previous images and the following images are predicted, at encoded this method.

The method is in no way based on its application a certain coding transformation, for example the Discrete Cosine Transformation (DCT), or on certain special procedures in the other Ver described above driving steps, limited.  

When coding, it happens that whole macroblocks, which an address is assigned to each of which the macro block is located within the encoded image, or just individual picture blocks are not encoded. These uncoded The decoding unit recognizes macroblocks (skipped macroblocks) usually by taking the difference of the address of the current or macro block and the previous macro block telt. If this difference is greater than zero, then one of the Difference corresponding number of uncoded macroblocks between between the current macroblock and the previous macro block.

Depending on the block-based coding method used the uncoded macroblocks are treated differently.

For example, the MPEG 1 method and the MPEG 2 method in P-pictures the motion vectors that the uncoded macroblocks are assigned the value zero pointed. With the H.261 standard, however, the movement vector of the macro preceding the uncoded macroblock blocks is assigned the value zero and the Prediction performed.

In the case of B pictures, the MPEG 1 process and the MPEG 2 Procedure the type of prediction from the previous Ma Kroblock taken over. The MPEG 1 process uses the so-called differential motion vector to zero ge puts. With the MPEG 2 method, the value of the movement vector from the value of a parameter "predictor" that matches the Motion vector value of the last encoded macroblock, predi graces.

In the known methods, the Images, i.e. a sequence of macro blocks during decoding the variable length decoding (VLD) fed. Exist subsequently no uncoded macroblocks, so both inverse quantization (IQ) and inverse coding  tion transformation (IDCT) as well as motion compensation (MC) processed in sequence. However, the episode has one uncoded macroblock, so the motion comp sation the uncoded macroblock like coded macroblocks processed. This means that a zero block is added to the Refe renzblock is added.

A disadvantage of this procedure is that these unnecessary additions cost unnecessary power dissipation.

Another disadvantage is that this approach takes considerably longer than the Ver invention drive as described below.

The invention is based on the problem of an arrangement and a method for coding and decoding a block based coding method that the in the previous described disadvantages of the known method avoided.

The problem is solved by the arrangements according to claim 1 and claim 2 and the method according to claim 3 and claim 4 solved.

In the method according to claim 3 it is checked whether the current macroblock is encoded or uncoded. Is he co the usual procedure for reconstruction of the macro block. However, if the macroblock is unco dated, it becomes a reference macroblock, including the corresponding one appropriate macroblock of the previous image , simply adopted as the current macroblock by the reference macroblock to the corresponding position of the pre seen image memory is copied. This can be parallel to that Coded macroblocks are processed, which means a considerable amount Liche time saving compared to the known method is enough.  

This time saving in decoding is special Significance, since it guarantees the decoding of an image certain real-time conditions only a predetermined one Time is available.

In the event that there are some macro blocks at the beginning of a picture processed and then a series of macroblocks that need increased effort to decode the macroblocks, may remain with the known method, se quantitative processing of all macroblocks, no longer enough enough time to decode the image reconstruction major macroblocks. This leads to a significant in loss of formation caused by the known method gently. To avoid this problem, one would be significant Lich more expensive and complex hardware necessary to get the real thing time condition to ensure that no loss of information occurs. Through the new process or the new arrangement will solve the problem of timing when decoding the images defused.

Furthermore, the additional unnecessary additions are ver avoided and a considerable saving in loss loss stung to carry out the procedure.

Furthermore, the dimensioning of the decoding unit or the coding unit significantly improved. The decoding unit or the coding unit can, for example, the processing direction of a macro block line can be dimensioned or on the processing of an entire image or a field. Dimensioning of the decoding unit or coding The processing of an entire image has a certain effect less design effort for the decoding unit or the Co the unit. This will also reduce the cost of Decoding unit required to carry out the method or coding unit reduced. This is achieved through the time saved by the inventive method  and the resulting lower demands on the hardware the decoding unit or the coding unit.

These advantages also apply to the method according to the patent claim 4.

Here it is again checked whether the current macroblock co dated or uncoded. If it is coded, the usual one Procedure for the reconstruction of the macro block leads. However, if the macroblock is uncoded, the Ma Kroblock marked as uncoded. The real verar processing, i.e. copying the reference block into the Image storage takes place only after the processing of a specified number of macroblocks.

For the arrangements according to claim 1 and according to the patent Claim 2 also apply the advantages described above of procedures.

Advantageous further developments result from the dependent ones Claims.

Further time saving as well as saving the loss required performance is achieved through further training of the process enough for everyone within each macroblock Image block of the macro block is checked whether the image block co dated or uncoded. For an uncoded picture block becomes a copy of the reference picture block of the uncoded picture blocks stored in the image memory.

Two exemplary embodiments are shown in the drawings following explained in more detail.

Show it

Fig. 1 is a sketch showing the known method of the prior art;

Fig. 2 is a sketch describing the principle of the method according to the invention;

Fig. 3 shows an arrangement which is used for carrying out the method;

Fig. 4 is a sketch in which method steps are described in detail, which are performed in the copying module;

Fig. 5 is a sketch in which an arrangement is described, which is used for performing the method according to claim 2;

Fig. 6 is a flowchart showing the individual procedural steps of the method according to claim 1;

Fig. 7 is a flowchart showing the individual procedural steps of the method according to claim 2;

Fig. 8 is a flowchart in which the further development of the method is described is also checked in which whether the image blocks of the current macro block is coded or uncoded;

Fig. 9 is a sketch showing two computers, wherein images are encoded in a first computer and transmitted to a second computer, where they are decoded.

Referring to Figs. 1 to 9 according to the invention are further illustrated Ver drive.

In Fig. 1, the known method is shown with a sketch.

Here is an image to be encoded or decoded divided into macro blocks MB, which have a predeterminable number Have image blocks BB.

The inventive methods are necessarily from a computer or special hardware, with which a coding unit or a decoding unit reali be settled.

In the two exemplary embodiments, two types of image blocks are distinguished, luminance image blocks Y _n ^m and chrominance image blocks C _r ^m , C _b ^m . Here, a macro block identifier m, which is a natural number, uniquely designates each macro block MB of a sequence of macro blocks MB. Depending on the block-based method used, different formats are used. A natural number n uniquely identifies each luminance image block Y _n ^m within a macro block.

This means a different division of image blocks ken BB within a macroblock MB.

In the exemplary embodiments, for easier illustration assumed a sequence of three macroblocks MB, one first macroblock, a second macroblock and a third Macroblock.

However, this in no way limits the generality Way.

In a coding unit, image areas of the to be coded Images to image blocks and several image blocks to macro blocks summarized. The image blocks have several pixels of the Image whose information, for example luminance information  mation or color information (chrominance information), coded becomes. For example, an image block has a size of 8 × 8 or 16 × 16 pixels.

A coding transfer is made to the individual macroblocks MB formation DCT applied, for example a Discrete Cosi nus transformation (DCT). The resulting transfor Mation coefficients are in a quantization stage Q quantized. The quantized transformation coefficients are subjected to a scanning process SV, for example the the so-called Zig-Zag scan method or the so-called Al ternate scan method. Then the scanned Transform coefficients a run length coding and a variable length coding was carried out.

The coded transformation coefficients and thus the co dated image blocks BB or macro blocks MB become a De coding unit transmitted over a channel where the macro blocks be decoded and the encoded image is reconstructed. It is also a reconstruction in the coding unit itself provided since only differential information is encoded. For this it is necessary also in the coding unit the coded picture to reconstruct.

From this it is clear that the method according to the invention both in the coding unit and in the decoding unit can be used.

During the decoding, the luminance picture blocks Y _n ^m as well as the chrominance picture blocks C _r ^m , C _b ^{m are} fed to the decoding unit. Method steps are now carried out in the decoding unit which are inverse to the method steps of the coding.

The decoding is a variable length decoding VLD, a run length decoding RLD, an inverse scan IS method, an inverse quantization IQ and an inverse  Coding transformation IDCT performed. Will continue a movement component to reconstruct the encoded image sation MC using an image memory FM leads.

When reconstructing the coded images in one back upward path in the coding unit are also those in the previous Gen described process steps for decoding leads to the variable length decoding VLD and the Run length decoding RLD.

The luminance image blocks Y _n ^m and the chrominance image blocks C _r ^m , C _b ^m are processed sequentially in the known method. This has the result that in the event that the sequence of macroblocks MB has an uncoded macroblock UMB or an uncoded image block UBB, the encoded macroblocks in the inverse coding transformation IDCT are already being processed, although initially with the motion compensation MC UMC uncoded macro blocks can be processed. This restriction means that the inverse coding transformation IDCT is only started with the processing of an uncoded macro block UMB following the coded macro block if the respective current data from the inverse coding transformation IDCT can also be processed with the motion compensation MC. As a result, when an uncoded macro block UMB or an uncoded image block UBB occurs, the inverse coding transformation IDCT is interrupted so that the motion compensation MC can be carried out for the uncoded macro block UMB or the uncoded image block UBB.

In each vertical time slot t _i , an image block BB is fed to a means shown in FIG. 1 for carrying out the method steps described above, and an image block BB is also removed. During each time slot t _i , in this exemplary embodiment in the unit for performing the variable length decoding, a maximum of one image block BB, in the unit for performing the inverse quantization IQ and inverse coding transformation IDCT, two image blocks BB and in the unit for performing the motion compensation MC processes six image blocks BB.

However, processing depends on the implementation, which every specialist is familiar with.

In the embodiment, it is assumed that the second Macroblock is UMB uncoded and that a second luminance block Y₁³ is an uncoded picture block UBB.

Waiting time steps Wt _i are shown in FIG. 1 for in the unit for carrying out the inverse quantization IQ and inverse coding transformation IDCT.

In Fig. 2, the solution principle of the United method according to the invention is outlined. Here, an additional copy module KM is provided. With the help of the copying module KM, in the case of uncoated macro blocks UMB or uncoded picture blocks UBB, simply reference macro blocks RMB or reference picture blocks RBB are copied from the temporally previous picture, which are stored in the picture memory FM. By copying and thus omitting the explicit calculation of the "new" picture blocks BB, which is carried out in parallel with the decoding of encoded macro blocks MB or picture blocks BB, the waiting time steps Wt _i , which are shown in FIG. 1, are avoided.

FIG. 3 shows what information is used to check whether a macroblock MB is encoded or not, and what information is used to reconstruct the image blocks BB. These parameters P are contained for each picture block BB or macro block MB and are transmitted together with this. These parameters P can thus be used directly in the decoding. The parameters P are, for example, a so-called Coded_Block_Pattern CBP, a macro block address ADR, an image type TYP, and movement information MI. The Coded_Block_Pattern CBP indicates whether the respective image block BB or macro block MB is coded or uncoded. The macro block address ADR indicates the address of the macro block MB within the image. The TYPE picture type indicates whether the coded picture is an I picture, a P picture or a B picture.

After the macroblocks MB have been received (using the methods in the decoding unit) or immediately after they have been subjected to the variable length coding VLC 600, the macroblocks MB are fed 601 to the variable length decoding VLD (see FIG. 6 ).

Then, for example, based on an address difference limit that is formed between the macroblock addresses two he successive coded macro blocks MB or also based on the Coded_Block_Pattern CBP or on another Parameters P with similar information content, depending on the use procedures, checks whether the macroblock MB an uncoded macroblock UMB or an encoded one Macroblock acts 602.

If the macroblock MB is encoded, the macroblock MB becomes the in the method steps described above for decoding of the macroblock MB: the run length decoding RLD 603, the inverse scan method IS 604, the inverse quanti IQ 605, the inverse coding transformation IDCT 606 and the motion compensation MC 607.

However, if the macroblock MB is uncoded, the required one is used Address information of the reference macroblock RMB of the previous 608 to the copying module KM, in which the Reference macroblock RMB is copied from the image memory FM again in the image memory RMB, but this time as the ak  current macroblock MB. This can be done in parallel with the editing of the encoded macroblocks.

Using the decoded macro blocks MB and the copied Ma In a final step 609, kroblock will decode that reconstructed picture.

This procedure described in the previous is shown in Fig. 4 for the MPEG 1 method.

In the MPEG 1 process, each macroblock MB has a move Assignment vector MV assigned in the header of each macro block MB is specified.

The motion vector MV becomes an uncoded macro block UMB assigned the value zero. It is therefore for the uncoded Macroblock UMB no coding transformation coefficients formed and transferred. The number in a row the uncoded macro blocks UMB is determined by the difference in the addresses of two successive encoded macroblocks fixed.

In a head evaluation HA, the motion vector MV for determines the current macroblock MB. This is done on a for those skilled in the art.

The value of the motion vector MV is, for example, in one Vector FIFO memory MVS stored. The value of the movement supply vector MV is in an address FIFO memory AS Address of the macroblock MB stored within the Picture indicates the number of the current macroblock MB. The Address of the respective macro block MB and the value of the move Delivery vectors MV are clearly assigned to one another.

Using an address list AL of the macro blocks MB is in ei nem Adress-Generator AG the physical address of the Ma kroblocks MB formed in the image memory FM, depending on  Image type, depending on whether it is the decode image is an I-picture, a P-picture or a B-picture.

If it is an I-picture, the address Generator AG the address of the macroblock MB in the new one (current) picture, because with this picture type the entire picture formation was encoded.

If it is a P-picture, the address Generator AG the address of the macro block MB of the reference ma kroblocks RMB in the previous picture, because at this image type the difference information is transmitted.

If it is a B-picture, it is in an intermediate stage ZS depending on the values of the motion vectors MV of the ent speaking reference blocks RMB of two successive I- Images or P-images that are used to reconstruct the B-image are used, the motion vector of the macro block MB des B-image determined by interpolation. Furthermore, the Pi xelinformation (luminance and chrominance of the respective Pi xels) also interpolated.

The detailed procedure here again depends on the coding method used and is known to the expert the known coding method known.

In the event that either a P-picture or a B-picture deco the reference macro blocks RMB, which are under the determined address are stored in the image memory FM, copied into a buffer and possibly a half Subjected to pixel filtering. Then he will now averaged data of the current macroblock MB in the image memory cher FM stored at the address of the current macro block MB.  

This is not necessary for an I-picture, because in this case the address determined by the address generator AG already represents the address of the current macroblock MB.

A second embodiment is shown in FIGS. 5 and 7.

Here the copy actions are not parallel to the decoding tion of the encoded macroblocks MB performed. The uncoded th macro blocks UMB are marked 701 and reconstruction data of the uncoded macro blocks UMB are in a recon structure memory RS saved.

Here, the value of the motion vector MV is, for example stored in the vector FIFO memory MVS. this happens however only for B-pictures to be decoded. The value of the movement supply vector MV is in an address FIFO memory AS Address of the macroblock MB stored within the Picture indicates the number of the current macroblock MB. The Address of the respective macro block MB and the value of the move supply vectors MV are clearly drawn towards each other in B-pictures arranges.

Only after the encoded macroblocks MB have been decoded the macro blocks UMB marked as not coded processed. The processing itself takes place in the Copy module KM 608 in the manner described above.

The serial processing can copy the Refe reference macroblocks RMB also in the means for motion estimation MC can be performed.

The remaining process steps correspond to the steps of the first embodiment.

Hybrid forms between these two "extreme cases" of the parallel len processing and the strictly separate processing of co  dated macro blocks MB and the uncoded macro blocks UMB are vorese in variants of the inventive method hen.

A further development of the method is shown in FIG. 8 using the example of the first exemplary embodiment. In a further step, it is checked for encoded macroblocks MB whether individual image blocks of the current macroblock MB are not coded 801. If uncoded image blocks UBB exist, the image blocks for the uncoded image blocks UBB as well as the uncoded macroblocks UMB in in the previous one the copying module KM formed from the corresponding reference image block RBB of the previous image.

In Fig. 9, an arrangement is shown in which the proced ren are used. In a first computer R1, the coding unit, an image is encoded in the manner described above. The image is recorded, for example, by a camera K.

The first computer R1 is coupled to a second computer R2 pelt. The first computer R1 transmits the encoded picture to the second computer R2, where the picture is decoded. The decoder te picture is now, for example, on a screen BS2 second computer R2 to a second viewer B2.

It is also contemplated that the encoded picture will not only transmitted but also decoded in the first computer R1 becomes what is necessary anyway with the block-based methods to determine the difference information actually transmitted tion, and on a screen BS1 of the first computer R1 egg is shown to the first viewer.  

The following publications have been cited in this document:
[1] D. Le Gall, MPEG: A Video Compression Standard for Multimedia Applications, Communications of the ACM, vol. 34, no. 4, pp. 47-58, April 1991, 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, 1991;
[3] GK Wallace, The JPEG Still Picture Compression Standard, Communications of the ACM, vol. 34, no. 4, pp. 31-44, April 1991, 1991.

Claims (5)

1. decoder for decoding images coded using a block-based coding method,

• a first means is provided for carrying out a variable length decoding (VLD) of a macroblock (MB) of an image (600) which has a predeterminable number of image blocks (BB),
• a control means (HA) is provided for checking whether the macroblock (MB) is uncoded (UMB),
• - in which a copy module (KM) is provided,
• a second means is provided for performing run length decoding (RLD),
• in which a third means is provided for carrying out an inverse scanning method (IS),
• a fourth means is provided for performing an inverse quantization (IQ),
• a fifth means is provided for carrying out an inverse coding transformation (IDCT),
• - In which a sixth means is provided for performing motion compensation (MC), and
• - In which an image memory (FM) is provided.

2. encoder for decoding images coded with a block-based coding method,

• in which a transformation unit is provided for performing a coding transformation (DCT),
• in which a quantization unit is provided for carrying out a quantization (Q),
• in which a scanning unit is provided for carrying out a scanning method (SV),
• a control means (HA) is provided for checking whether the macroblock (MB) is uncoded (UMB),
• - in which a copy module (KM) is provided,
• in which a means is provided for carrying out an inverse scanning method (IS),
• - in which a means is provided for performing an inverse quantization (IQ),
• in which a means is provided for carrying out an inverse coding transformation (IDCT),
• - In which a means is provided for performing motion compensation (MC), and
• - In which an image memory (FM) is provided.

3. Method for decoding images coded with a block-based coding method,

• variable length decoding (VLD) (601) is carried out for a macroblock (MB) of an image (600) which has a predeterminable number of image blocks (BB),
• - which checks whether the macroblock (MB) is uncoded (SMB) (602)
• - In the event that the macroblock (MB) is coded, the following steps are provided:
  • a run length decoding (RLD) is carried out for the macroblock (MB) (603),
  • an inverse scanning process (IS) is carried out for the macroblock (MB) (604),
  • - an inverse quantization (IQ) is carried out for the macroblock (MB) (605),
  • an inverse coding transformation (IDCT) is carried out for the macroblock (MB) (606),
  • motion compensation (MC) is carried out for the macroblock (MB) (607),
  • - in which the macroblock (MB) is saved, and
• - In the event that the macroblock (MB) is uncoded, a copy of a reference macroblock (RM) of the uncoded th macroblock (MB) is stored in an image memory (FM) (608), and
• - in which the decoded macroblock (MB) or the copy of the reference macroblock (RM) represent a reconstructed macroblock (RMB) (609).

4. Method for decoding images coded with a block-based coding method,

• variable length decoding (VLD) (601) is carried out for a macroblock (MB) of an image (600) which has a predeterminable number of image blocks (BB),
• - which checks whether the macroblock (MB) is uncoded (SMB) (602),
• - In the event that the macroblock (MB) is coded, the following steps are provided:
  • a run length decoding (RLD) is carried out for the macroblock (MB) (603),
  • an inverse scanning process (IS) is carried out for the macroblock (MB) (604),
  • - an inverse quantization (IQ) is carried out for the macroblock (MB) (605),
  • an inverse coding transformation (IDCT) is carried out for the macroblock (MB) (606),
  • motion compensation (MC) is carried out for the macroblock (MB) (607),
  • - in which the macroblock (MB) is saved, and
• - In the event that the macroblock (MB) is uncoded, the macroblock (MB) is marked as uncoded (701) f
• - In which, after the decoding of a predeterminable number of macroblocks (MB) for the macroblocks marked as uncoded, (MB) a copy of a reference macroblock (RM) of the respective uncoded macroblock (MB) is stored in an image memory (FM) (608 ), and
• - in which the decoded macroblock (MB) or the copy of the reference macroblock (RM) represent a reconstructed macroblock (RMB) (609).

5. The method according to claim 3 or 4,

• - In which for each picture block (BB) of the macro block (MB) ge is checked whether an image block (BB) is uncoded, and
• - In which, in the event that the picture block (BB) is uncoded, a copy of a reference picture block (RB) of the uncoded picture block (BB) is stored in the picture memory (FM).