DE19617171C2 - Video decoder and corresponding method - Google Patents

Description

The invention relates to a video decoder and a corresponding procedure.

The so-called MPEG standard has been established for the coding of video images, which combines the principles of transformation and prediction coding to form what is known as hybrid coding. It provides three types of encoded images that differ in the type of data compression:

• 1. I-pictures (intra-coded pictures): The coding of an I- Image is independent of other images. An I picture serves to synchronize the transmission of several subsequent video images and their coding uses the local correlation of the image data of the I-image. The Coding is preferably a discrete cosine Transformation, DCT followed by quantization and Weighting of the transformation coefficients with final Entropy coding.
• 2. P-pictures (predicted pictures): their coding depends from a previous I or P picture (Forward prediction). P pictures become one Motion estimation related to the previous picture subject (motion-compensated prediction). Subsequently the intra-picture coding (e.g. to the above for an I- Described way) to the temporal prediction error applied.
• 3. B-pictures (bidirectionally predicted pictures): With these finds both a temporally motion-compensated prediction regarding a previous one as well as regarding of a temporally following P or I picture instead. Man also speaks of "motion-compensated interpolation". The  Expressions (in time) "subsequent" or "previous" do not refer to the order of transmission of the images in the data stream of the encoded images, but on their recording / playback order. Pictures, the base a temporal prediction for the coding of another Images are also referred to as "support images". As well Like P-pictures, B-pictures are also quantized Coded transformation coefficients of a difference image.

It is also known, in particular for display by means of TVs according to the PAL or NTSC standard, playback of video images not as a so-called full screen but as make two consecutive so-called fields ("interlaced format"). A full screen shows the whole video image to be displayed and thus all lines of this picture. Full images are also called "frames" designated. In contrast, each of the two fields points ("field") only half of all lines of one Video image on. In the so-called "top field" or the first Field is all even lines, at "Bottom field" or second field around all odd numbers Lines of the associated full screen.

An MPEG decoder does not only have to encode the video images decode in the correct order, but also if e.g. B. a representation for televisions according to the PAL or NTSC standard is provided for a provision of this provide the necessary two fields. The transferred However, video images can be either frame or field encoded his. In the former case, the transmitted video images are Full images, in the second case fields, where each two of these again form a full image to be displayed.

MPEG decoders have storage means for storing previously decoded support images. Saving this I or P Images are necessary because they are based on a temporal Prediction of a P or B picture are and therefore for it  Decoding will be needed. Should z. B. one from the decoder received, encoded B-picture is processed, finds one Decoding the B-picture with the help of the two for its bidirectional prediction used by the decoder previously received and saved support images instead.

Furthermore, it is known that the decoding of a B- Support images not decoded, but in to store the coded form in the decoder in order to Save space. However, this method has the Disadvantage that the decoder can handle Data throughput increased significantly. The same I and P pictures serve namely generally for decoding a large number of B- Pictures so that they are decoded accordingly frequently have to. It is even more important, however, that according to MPEG a block-based coding is provided, the for the Prediction areas used in the Support images often deviate from the block boundaries. Therefore have to decode a block from a B picture u. U. ever four blocks of the two coded stored supporting images be decoded. It becomes an eight-fold data throughput compared to the case of storing decoded supporting images necessary. This eightfold amount of data must still be in the same unit of time, which is total for the Decoding of the predicted encoded image is available stands. The performance requirements of the Decoders are therefore considerably larger than in the case of decoded saved support images.

In conventional MPEG decoders, a memory is also included tel provided in which the decoded B-picture is stored ("Output frame buffer"). This storage medium is in the case a frame-coded B picture in the conventional decoders necessary because, as already mentioned, by the decoder from the Full screen two fields are to be generated, which are used for Display of the full picture in interlaced format one after the other must be present at an output of the decoder. The starting  In the case of field-coded B-pictures, however, the frame buffer also necessary for another reason: some Applications may be necessary (e.g. for Adaptation of different frame rates between Recording system and playback system) that for display not just two fields of a full screen be, but that according to the representation of the second Field again the first field is repeated.

The invention has for its object a video decoder to be specified at which the memory requirement is low.

This task is performed by a decoder according to claims 1 and solved a decoding method according to claim 7.

According to the invention, the first image is not decoded (as in the prior art), but in coded Save shape in the decoder. To generate the for the The necessary fields are reproduced at least double decoding. After each decoding there is one Output of the decoding result to an output of the Decoder to it (for example by means of a television set) to play. An output frame buffer, as in the prior art Technology, is not necessary. To save the encoded The first image requires less storage space than if it would be stored decoded, as in the prior art the case is. The invention is not in the case of MPEG Limited decoders. It refers to wherever from a first image for reproducing multiple information must be generated. In contrast to the state of the art the stored, encoded first image for generating the different information (e.g. two fields a full picture) each time decoded (the picture may only partially and not completely decoded each time got to). In the prior art, there is first a complete decoding of the image which is stored and is subsequently output multiple times (e.g. output of the  two fields of a decoded frame). The The invention is also applicable to decoding methods in which no time prediction is made and none Support images are provided.

A further development of the invention relates to a video decoder according to the MPEG standard and provides that the first picture is temporally predicated on a second picture, but itself is not required for decoding another picture (i.e. it does not serve to predict another person's time Picture and therefore does not have to be used for decoding another Image can be decoded). It can be such for example, the first picture is a P or B picture.

Compared to the state of the art in which storage the coded support images (namely I and / or P images) is provided, the invention has the advantage that a much smaller amount of data has to be decoded. this has The following reasons: In the state of the art mentioned before the decoding of a B picture the coded stored I- and P-pictures, which are its temporal prediction underlying, are first decoded. This decoding the I and P pictures must be for each B picture to be decoded be made again. This results in a Multiplication of the data to be decoded by the decoder.

The invention, however, can in the worst case result in a doubling of the amount of data to be decoded, namely, when the case of a frame-coded B picture as first picture is considered, in which according to the invention first decoding to generate the first field and a second decoding to generate the second field from the coded frame. In an advantageous way can do this doubling of the amount of data to be decoded to be avoided, because to obtain a field from a encoded full screen mostly limited to Decoding the relevant for the field to be generated  Lines can be done. This applies, for example, to the Motion compensation, but also for most cases of inverse quantization and inverse DCT.

The invention is tert tert in the following with reference to the drawings. Figs. 1, 2 and 4 show different exporting approximately embodiments of the invention. Fig. 3 shows the structure of a conventional MPEG decoder, which is first explained here.

An encoded second image I, an encoded third image P and an encoded first image B are successively applied to an input In of the conventional decoder for video images from FIG . This corresponds to the order in which the data is transmitted, not the order in which it is recorded or played back, and depends on the type of coding. Assume that the first picture B is a B picture, the second picture I is an I picture and the third picture P is a P picture.

The coded image currently in each case is temporarily stored in an input buffer 1 and then subjected to entropy decoding 2 , inverse quantization and inverse DCT (discrete cosine transformation) 3 and, if the respective image is temporally predicted, motion compensation 4 .

An output Out of the decoder is preceded by a post-processing unit 5 (post processing) which, in the present case, assumed to be frame-coded pictures I, P, B as fields I _T , I _B or B _T , B _B or P _T , P _B in interlaced format. The second picture I and the third picture P, which are necessary for the decoding of the first picture B as supporting pictures, are decoded and stored in a second storage means S2, while the decoded first picture B in an output frame buffer S for the subsequent decomposition into the two fields B _T , B _{B is} stored.

The pictures decoded by the decoder at the output are Out then for playback to a playback device D giveable.

In the prior art it can also be provided that Support images I, P in the coded state and save them for each decoding to be performed one on them in time to decode predicted B-picture.

Fig. 1 shows a first embodiment of the invention. As already known in the prior art decoder in Fig. 3, the decoder according to the invention a Eingangspuf fer 1, an entropy decoder 2, a unit 3 for imple tion of the inverse quantization and inverse DCT, a means for motion compensation 4 and a post-processing unit 5 ( Post Processing) and a second storage means S2 for storing a decoded second picture I and a third picture P, which serve as supporting pictures for a subsequent decoding. In this embodiment, it is assumed that the second image I is an I-image and the third image P is a P-image that is temporally predicted with respect to the second image I.

At an input In of the decoder, the co dated second picture I, the encoded third picture P and a first picture B (in this case a B picture which is related to of the two previous supporting images I, P in time is predicted bidirectionally). During the second Picture I and the third picture P are decoded in a known manner and are stored in the second storage means S2, the first Image B first in a first storage means S1 in stored in coded form.  

At an output Out of the decoder, the fields I _T , I _B , B _T , B _B , P _T , P _{B of} the I-picture I, the B-picture B and the P-picture P are given, as they are for a subsequent illustration , e.g. B. on a television set, in interlaced format in this order by the decoder. The transmission of the coded pictures I, P, B in a different order than is necessary for playback (at the output Out of the decoder) follows from the principle of bidirectional prediction and is known from the prior art.

The first picture B, which is frame-coded in this exemplary embodiment, that is to say represents a full picture, which is stored in the first storage means S1, is now decoded once in each case to generate each of the two fields B _T , B _B by means of the supporting pictures I, P. Under certain circumstances, for example, triple decoding may also be necessary if, after the generation of the second field B _B , the first field B _T is again necessary for the reproduction of the transmitted video images (in FIG. 1 the second transmission of the first field was shown B _T at the output Out indicated by putting in brackets).

Of FIG. 1 it can be seen that between the decoding stages 1 to 4 and the output Out of the decoder, no further storage means for storing the decoded first image B or fields B _T produced therefrom, is provided _B B, so that the latter without intermediate storage directly can be given at the output Out. The output frame buffer S of the prior art is therefore omitted, see FIG. 3.

According to the MPEG standard, it is customary to split images into so-called macro blocks (e.g. six blocks of 8 × 8 pixels). The result of this is that so-called macroblock lines are only used during decoding - these are groups of e.g. B. 16 image lines each - instead of the individual image lines. However, the output from the decoder for subsequent display must be done line by line. In order to convert this "block-oriented" format into a "line-oriented" format, a small buffer must be provided for the application of the invention at the location where the output frame buffer S ( FIG. 3) is arranged in the prior art. In contrast to the prior art, however, the buffer is only used to hold one decoded macro block line (e.g. 16 image lines) and not the entire field or frame to be decoded (e.g. 288 or 576 image lines) and is therefore large smaller than an output frame buffer. After the entire macroblock line has been decoded and stored in the buffer, the line-by-line output required for the reproduction can take place. In order to achieve an uninterrupted output of decoded picture lines, it makes sense to dimension the buffer so that it can hold two macro block lines (for example a total of 32 picture lines). After a macro block line has been decoded, it can then be output line by line, while at the same time another macro block line is being decoded and the data generated in this way are also continuously stored in the buffer memory without the previous macro block line being overwritten.

The first storage means S1 can advantageously be part of the input buffer 1 , so that it only has to be dimensioned larger compared to the prior art in order to be able to record the images arriving at the input In of the decoder in addition to the first image B to be stored . In contrast to the prior art ( FIG. 3), no output frame buffer S is necessary in the invention, in which the decoded first picture B is stored. The first storage means S1 (plus the possibly required buffer, already mentioned in the previous paragraph) can be dimensioned smaller than the output frame buffer S in the prior art.

Advantageously, the fields B _T , B _B can be generated from the first frame B in such a way that in both cases not the entire first frame B is decoded, but only the lines necessary for the fields. As a result, the amount of data to be decoded can be significantly reduced.

As an alternative to the illustrated embodiment, it can be provided, in addition to the first image B also Support pictures I, P are not decoded, but also coded save. Then there is total within the decoder an even lower memory requirement, although the Decoding effort per unit of time increases.

Fig. 2 shows a second embodiment of the invention, which differs from that in Fig. 1 in that the first image B (also in this case a B-picture) is not a frame, but a first field B _T , which with egg nem additionally transmitted second field B _{B form} a field-coded frame, in contrast to the frame-coded B-picture B in FIG. 1. Both fields B _T , B _B are B-pictures and are temporally predicted with regard to the supporting pictures I, P. .

At the decoder in Fig. 2 illustrated the first field B _T is stored encoded in the first storage means S1, as it as already explained, may be necessary, turn output the first sub-image B _T after the output of the second field B _B at the output Out. For this purpose, the second picture I received first (an I picture) is decoded and stored in the second storage means S2, simultaneously split into two fields I _T , I _B and given in this order to the output Out. The third picture P next to the input In is decoded and likewise stored in the second storage means S2. The coded first sub-picture B _T then received at the input In is stored in the first storage means S1 and simultaneously decoded with the aid of the second image I and the third image P, which are stored in the second storage means S2, and then passed to the output Out. Then, with the exception of the storage in the first storage means S1, the same occurs with the second field B _B. Then the first field B _T stored in the first storage means S1 is decoded and output, whereupon the third image P stored in the second storage means S2 is finally broken down into fields P _T , P _B and output.

It is pointed out again that in block-oriented methods such as MPEG, a relatively small buffer for accommodating two macroblock lines must be arranged between the motion compensation 4 and the postprocessing unit 5 in order to be able to output individual image lines instead of the decoded macroblock lines.

In alternative embodiments of the invention, naturally Lich also be provided as the second image I no I-picture but also to provide a P-picture, since it is general is known, B-pictures both with regard to I-pictures and P- Predict images. The same applies to the case that both support images I to be stored in the first storage means Pictures are.

Furthermore, it is possible that the second image I (as an I or P image) is stored as the only supporting image in the second storage means S2 (this can then be dimensioned correspondingly smaller than in the exemplary embodiments in FIGS . 1 and 2) and that the first picture B is a forward-predicted P-picture, which was coded with respect to the second picture I, but itself is not used for prediction (this does not correspond to the MPEG standard, however).

It is of course also possible that the coded pictures I and P, which are assumed to be frames in FIGS. 1 and 2, are fields or are coded as fields, but are stored in the second storage means as frames.

Fig. 4 shows a further embodiment of the invention. This is an MPEG decoder, but the invention is of course not limited to such. In Fig. 4 are at the input in succession except two support frames I, P, two B-pictures, namely a first image B, and a fourth image B '. It is assumed that all pictures I, P, B, B 'are frame-coded, that is to say frames. In this embodiment, the two B-pictures B, B 'are stored in the first storage means S1. By multiple decoding of these two pictures B, B 'it is then possible, for example, to generate the fields shown at the output Out in the order given. In the present case, the first picture B is first decoded twice, then the fourth picture B 'is decoded twice and then the first picture B is decoded again twice in order to generate the two respective fields. Other chronological sequences can also be achieved, which are dependent on the form in which a reproduction by means of the reproduction device D is to take place.

In Fig. 4 of the already mentioned several times latch Z between the motion compensation 4 and the post-processing unit 5 is located. It is only necessary for block-oriented coding methods and can usefully include two macro block lines with an MPEG decoder. It is thus considerably smaller than an output frame buffer S according to FIG. 3. If the coding or decoding is not block-based but line-oriented, the buffer store Z is not necessary. The decoded images can then be output directly line by line.

It is possible that the decoder according to the invention is a component of a corresponding encoder in which the Motion estimation for the coding of the data to be transmitted  Video images is performed. MPEG encoders included basically also a decoder for the temporal Prediction.

The invention is particularly advantageously applicable to all video images that are not the basis for a temporal prediction of another image. This applies not only to the first images B shown in the two exemplary embodiments of FIGS . 1 and 2, which are assumed to be predicted bidirectionally as B images, but also (apart from unidirectionally predicted P images) also non-temporally predicted images, such as e.g. . B. I-pictures.

Claims (8)

1. Video decoder with the following features:

• 1. it has a first storage means (S1) for storing a coded first picture (B),
• 2. the stored, coded first picture (B) is decoded several times,
• 3. The results of the decodings can each be given to an output (OUT) of the decoder for reproduction of the decoded first picture (B) by means of a reproduction device (D).

2. Video decoder according to claim 1 with the following features:

• 1. the coded first picture (B) is temporally predicted with respect to a second picture (I), but does not itself serve to predict the time of another coded picture,
• 2. the decoder has a second storage means (S2) for storing the second image (I),
• 3. The stored, coded first picture (B) can be decoded several times with the aid of the stored second picture (I).

3. Video decoder according to one of the preceding claims with the following features:

• 1. the encoded first picture (B) is a full picture,
• 2. A first field (B _T ) can be generated from the stored, coded first picture (B) by a first decoding and a second field (B _B ) by a second decoding.

4. Video decoder according to claim 3, in which after the generation of the second field (B _B ) by a third decoding of the stored, coded first picture (B) in turn the first field (B _T ) is decodable. 5. Video decoder according to one of claims 1 or 2 with the following features:

• 1. the encoded first picture (B) is a first field (B _T ),
• 2. is in succession
  • 1. the stored, coded first field (B _T ),
  • 2. a coded second field (B _B ) and
  • 3. the stored, coded first field (B _T ) can be decoded again.

6. Video decoder according to one of the preceding claims with the following features:

• 1. in the first storage means (S1), in addition to the first picture (B), a fourth picture (B ') can be stored, which can then also be decoded several times,
• 2. The results of the decoding of the first (B) and fourth (B ') picture can be given to the output (Out).

7. A method for decoding a coded first picture (B) with the following features:

• 1. the encoded first picture (B) is stored and then decoded several times,
• 2. The results of the decodings can each be given to an output (OUT) of the decoder for reproduction of the decoded first picture (B) by means of a reproduction device (D).

8. The method according to claim 7 with the following features:

• 1. the coded first picture (B) is temporally predicted with respect to a second picture (I), but does not itself serve to predict the time of another coded picture,
• 2. the second image (I) is saved,
• 3. The stored, coded first picture (B) can be decoded several times with the aid of the stored second picture (I).