JP2006503475A - Drift-free video encoding and decoding method and corresponding apparatus - Google Patents

Abstract

Three-dimensional (3D) subband coding schemes use motion compensation in their temporal filtering stage. Unfortunately, this procedure introduces two problems. (A) MC is applied at full resolution, and drift appears when decoding at low resolution. (B) All motion vectors predicted at full resolution are transmitted, which is a waste of the number of bits. According to the present invention, a low-resolution sequence is obtained by first generating a low-resolution frame sequence from the original input frame sequence by wavelet decomposition and performing motion compensation space-time analysis on the low-resolution frame. It is done. A full-resolution group-of-frame motion-compensated space-time analysis is then performed, with the corresponding low-frequency subbands of the generated low-resolution sequence at the respective time-resolving level, the low-frequency for the decomposition. Subbands are replaced last. The modified sequence thus obtained is finally encoded. Thanks to this approach, good behavior at low resolution is maintained while approaching the performance of a classic 3D subband codec at full resolution.

Description

  The present invention relates to an encoding method for compression of an original video sequence divided into successive group of frames (GOF) and corresponding decoding. The present invention also relates to a corresponding encoding device and decoding device.

  Internet growth and advances in multimedia technology have enabled new applications and services for video compression. Many of them not only require coding efficiency, but also require extended functionality and flexibility to adapt to changing network conditions and terminal capabilities, and scalability meets these demands. Is. Current video compression standards may be based on a hybrid DCT (Discrete Cosine Transform) prediction structure and already include some scalability features. The hybrid structure is based on a prediction scheme in which each frame is temporally predicted from a given reference frame (prediction options are forward for P frames or bidirectional for B frames) and are then obtained The prediction error is spatially transformed to obtain the advantage of spatial redundancy (a two-dimensional DCT transform is used in the standard scheme). Scalability is achieved thanks to a further enhancement layer.

  Alternatively, three-dimensional (3D) subband video coding techniques generate a single, embedded bitstream with full scalability. This 3D subband video coding technique relies on space-time filtering that allows reconstruction at the desired spatial resolution or frame rate. Such an approach has been proposed, for example, in the literature “Thee-dimensional subband coding of video”, C. Podilchuk and al., IEEE Transaction on Image Processing, vol. 4, No. 2, February 1995, pp. 125-139. The group of frames (GOP) is processed as a three-dimensional (2D + t, or 3D) structure and is spatio-temporal filtered to agglomerate energy at low frequencies. Including motion compensation in this scheme to improve efficiency).

  FIG. 1 shows the 3D subband structure obtained by such an approach, where the exemplified 3D wavelet decomposition with motion compensation is applied to the group of frames, and this current GOF is a sequence with large motion. Is first motion compensated (MC) and time filtered (TF) using Haar wavelets (dashed arrows correspond to high pass time filtering, other arrows are low pass time filtered) Corresponding to). After the motion compensation operation and temporal filtering operation, each temporal subband is spatially decomposed into space-time subbands, so that a final 3D wavelet representation of the original GOF is obtained, and the three decomposition stages are: Shown in the example of FIG. 1 (L and H = first stage; LL and LH = second stage; LLL and LLH = third stage). The known SPIHT algorithm is extended from 2D to 3D and is selected to efficiently encode the final coefficient bitplane with respect to the space-time decomposition structure.

When implemented, this 3D subband structure applies the original full resolution motion compensation (MC) space-time analysis at the encoder side. Spatial scalability is achieved by removing the best spatial subbands for decomposition. However, when motion compensation is used in a 3D analysis scheme, the method does not allow complete reconstruction of video sequences at low resolution, even at very high bit rates. This phenomenon is referred to as drift in the following description and reduces the visual quality of the scalable solution compared to direct encoding at the intended final display size. As explained in the document “Multi-scale video compression using wavelet transform and motion compensation”, PY Cheng and al., Proceeding of the International Conference on Image Processing (ICIP95), Vol. 1, 1995, pp. 606-609. The drift results from the order of wavelet transform and motion compensation that are not interchangeable. When spatial scalability is enabled at the decoder side, spatial sub-band is skipped regarding the highest decomposition is performed at the encoder side, thereby, possible reconstruction or synthesis of lower resolution versions a _d of the original frame A To do. For such synthesis, the following operations are applied.

ここで、ＤＷＴ_L（空間領域における離散コサイン変換）は、３Ｄ分析におけるのと同じウェーブレットフィルタを使用して解像度のダウンサンプリングを示している。パーフェクト・スケーラブル・ソリューションでは、以下を有することを望む。

Here, DWT _L (discrete cosine transform in the spatial domain) shows resolution downsampling using the same wavelet filter as in 3D analysis. In a perfect scalable solution we want to have:

したがって、式（１）の残りの部分はドリフトに対応する。ＭＣが適用されない場合、ドリフトが除かれる。同じ現象は、固有の動きベクトルがフレームに適用される場合に（画像の境界を除いて）発生する。さらに、ＭＣが良好な符号化効率を達成するために回避不可能であることが知られており、固有の全体的な動きの可能性は、以下のパラグラフでこの特定のケースを除くために十分に小さい。

Therefore, the remaining part of equation (1) corresponds to drift. If MC is not applied, drift is eliminated. The same phenomenon occurs (except for image boundaries) when a unique motion vector is applied to the frame. In addition, MC is known to be unavoidable to achieve good coding efficiency, and the inherent overall motion potential is sufficient to exclude this particular case in the following paragraphs Small.

  Authors such as JWWoods et al. Have written the document “A resolution and frame-rate scalable subband / wavelet video coder”, IEEE Transaction on Circuits and Systems for Video Technology, vol.1, No.9, September 2001, pp.1035- 1044 has already proposed a technical solution to eliminate this drift. However, in such a document, the described scheme includes sending extra information in the bitstream in addition to being quite complex, thus wasting some number of bits. The solution described in the previously cited document "Multi-scale video compression ..." avoids this bottleneck, but operates in a predictive manner and cannot be replaced by a 3D subband codec.

Next, European Patent Application No. 02290155.7 (PHFR020002) filed on January 22, 2002 is a solution that avoids these problems, and the video coding method is a continuous group of frames (GOF). Proposed a solution according to including the following steps used for compression of the original video sequence divided into:
(1) A step of generating a low-resolution sequence including a continuous low-resolution GOF from the original video sequence by wavelet decomposition.
(2) performing a low resolution decomposition on the low resolution low resolution sequence by motion compensated space-time analysis of each low resolution GOF;
(3) generating a full resolution sequence from the low resolution decomposition by fixing high frequency spatial subbands resulting from wavelet decomposition to the low resolution decomposition;
(4) encoding the full resolution sequence and motion vectors generated during motion compensated space-time analysis to generate an output encoded bitstream;

  Such a solution that preserves the overall structure of the decomposition tree in 3DS analysis and does not send extra information to compensate for drift effects (only the decomposition / reconstruction mechanism changes) is the encoding scheme of FIG. It is recalled again in a more detailed manner with reference to the motion compensation time analysis at the lowest resolution illustrated in FIG.

Two main steps are provided. (A) A motion compensation step at the lowest resolution. (B) High spatial subband encoding step. First, motion compensation (MC) is applied at this level to avoid drift at lower resolutions. As a result, the GOF at full resolution (21 in FIG. 2) is first downsized (this step is indicated by reference d in FIG. 3 and corresponds to steps 22 and 23 in FIG. 2) and then normally The 3D subband MC decomposition scheme is applied to the downsized GOF instead of the full size GOF as shown in FIG. 3, and is illustrated by step 24 in FIG. In FIG. 3, the time subbands (L _{0, d} , H _{0, d} ) and (L _{1, d} , H _{1, d} ) are determined according to a known lifting scheme (H is first determined from A and B, L is then determined from A and H), the dashed arrow corresponds to high-pass time filtering, the solid arrow corresponds to low-pass filtering, (A _{0, d} , A _{1, d} , A _{2 , d} , A _{3, d} ) between the low frequency spatial subbands A of the frames of the sequence referenced by A _{, d} , or between the low frequency temporal subbands L referenced by L _{0, d} and L _{1, d} , Corresponding to motion compensation (a side effect of this method is a reduction in the amount of motion vectors to be transmitted in the bitstream, thereby saving the number of bits for texture coding). Entropy coder based on tree subbands (eg “Low bit-rate scalable video coding with 3D set partitioning in hierarchical trees (3D-SPIHT)”, BJKim and al. IEEE Transaction on Circuits Systems for Video Technology, vol.10, No. 8, December 2000, pp. 1374-1387, high space allowing full resolution reconstruction prior to transmission to eg 3D-SPIHT encoder 27 as shown in FIG. Place the subband (step 26 in FIG. 2). The final tree structure is described in “A fully scalable 3D sub-band video codec”, IEEE Conference on Image Processing (ICIP2001), vol.2, pp.1017-1020, Thessaloniki, Greece, October 7-10, 2001. Such a 3D subband codec structure looks very similar, so a tree-based entropy coder can be applied to it without limitation. In the coding scheme of FIG. 2, the reference codes (for the full resolution sequence 21) are as follows:

FRS: Full resolution series 21
WD: Wavelet decomposition 22
LRS: Low resolution series 23
MC-3DSA: Motion compensated 3D subband analysis 24
LRD: Low resolution decomposition (251)
HS: High subband 26
U-HFSS: Combined three high-frequency spatial subbands of the frame
FR-3D-SPIHT: Full resolution 3D SPIHT27
OCB: Output Encoded Bitstream The corresponding encoding scheme shown in FIG. 4 is symmetric with respect to this encoder (in FIG. 4, further reference symbols are as follows).

FR-3D-SPIHT: Decoding step 41
MC-3DSS: Motion compensated 3D subband synthesis 43
HSS: High subband separation 44
FRR: Full resolution reconstruction of full resolution series 45)
In order to enable spatial scalability, the high frequency spatial subbands must be cut as in the normal version of the 3D subband codec, and the decoding scheme of FIG. Shows you can get to.
Then, to encode the high spatial subband, two main solutions are proposed, the first solution is without MC and the second solution has MC.

  In the first solution, the high subband corresponds to the high frequency space subband of the original (full resolution) frame of the GOF in wavelet decomposition. These subbands allow reconstruction at full resolution on the decoding side. Indeed, the frame can be decoded at a low resolution. However, these frames correspond to low spatial subbands in the wavelet analysis of the original frame. Therefore, it is necessary to combine low resolution frames and corresponding high subbands, applying wavelet synthesis to obtain full resolution frames, thereby optimizing the 3D-SPIHT encoder. In the MC scheme for 3D subband encoders, the low time subband is always visible in one of the original frames of the GOF. As a matter of fact,

したがってＬはＡのように見える。結果的に、Ａの高い空間サブバンドは、Ｌに対応する低い解像度の分解で置き換えられる。（前方向ＭＣのケースにおける高い空間サブバンドを再オーダリングする）このアプローチは、図５に例示されており、ｊｔは、時間分解レベルを示しており（０はフルフレームレートについてであり、ｊｔ＿ｍａｘは最も低いフレームレートについてである）、ｎｆは、時間レベルｊｔでのサブバンドインデックスであり、ＤＷＴ_Hは、高周波ウェーブレットフィルタを示し、係数ｃ_jtは、乗算係数であり、ＯＦ，ＬＲＦ，ＴＳは、参照符号０〜３で参照されるオリジナルフレーム、参照符号００〜０３で参照される低解像度フレーム、及び送信されたサブバンドをそれぞれ示している。

Thus L looks like A. As a result, A's high spatial subband is replaced with a low resolution decomposition corresponding to L. This approach (reordering high spatial subbands in the forward MC case) is illustrated in FIG. 5 where jt indicates the time resolution level (0 is for full frame rate and jt_max is Nf is the subband index at time level jt, DWT _H denotes a high frequency wavelet filter, coefficient c _jt is a multiplication coefficient, and OF, LRF, TS are An original frame referred to by reference numerals 0 to 3, a low resolution frame referred to by reference numerals 00 to 03, and a transmitted subband are shown.

  In the second solution, using MC in each subband does not allow drift-free reconfiguration, so it is also possible to partially use MC to form higher spatial subbands. Each resolution can be reconstructed. Instead of directly using the high frequency space subbands of the wavelet decomposition, wavelet decomposition is performed on the prediction error obtained from the MC performed at full resolution, for example, reusing low resolution motion vectors.

  The object of the present invention is to improve the previously described solution by retaining its good behavior at low resolution while approaching the performance of a classic 3D subband codec at full resolution.

For the above purpose, the present invention relates to a video coding method for compression of an original video sequence divided into consecutive group of frames (GOF), which method comprises the following steps: ing.
(1) A step of generating a low-resolution frame sequence organized into a continuous low-resolution GOF from a full-resolution frame of an original video sequence by wavelet decomposition.
(2) A step of performing motion compensated spatio-temporal analysis on each low resolution GOF of the low resolution frame sequence to obtain a low resolution sequence.
(3) performing a motion resolution space-time analysis of each full resolution GOF of the original video sequence.
(4) At each time resolution level, replacing the low frequency subband of the decomposition with the corresponding space-time subband of the low resolution sequence.
(5) For the output encoded bitstream, encoding the modified sequence thus obtained and the motion vectors generated during the motion-compensated space-time analysis of the respective full resolution GOF.

The present invention also relates to a video decoding method corresponding to the previously defined video encoding method, and to a corresponding video encoding device and video decoding device.
The invention will now be described in more detail with reference to the accompanying drawings.

  With respect to the solutions described above, the present invention will now be described with reference to its basic steps. (A) Motion compensation at the lowest resolution (this first step, motion compensation (MC) is actually exactly equivalent to that described in the case of the previous solution. Using a spatial wavelet filter Downsize the GOF and then apply the normal 3D subband MC decomposition scheme to this downsized GOF), (b) High spatial subband coding.

  The main difference from the previous solution is in the second step, whose principle is to inject low spatial resolution analysis at each resolution level into those of full resolution analysis. Thus, it is possible to reconstruct the original frame at the decoder side while performing real time filtering (for high frequency spatial subbands, not intra coding or prediction errors as in the previous solution).

  The following equations illustrate the mechanism in a more detailed manner. As before, the first temporal analysis is performed at low resolution, which are represented by equations (4) and (5). There is a case.

次の表記による。Ａ＝参照フレーム、Ｂ＝現在のフレーム、ＤＷＴ＝離散ウェーブレット変換。Ａ_d＝フレームＡのＤＷＴの低周波空間サブバンド、すなわちフレームＡの低い空間解像度バージョン。Ｂ_d＝フレームＢのＤＷＴの低周波空間サブバンド、すなわちフレームＢの低い空間解像度バージョン。Ｈ＝低い空間解像度での高周波時間サブバンド。Ｌ＝低い空間解像度での低周波時間サブバンド。ＭＣ_down＝低下解像度（すなわちサブサンプルされた）フレームで実行された動き補償。ＭＣ^-1＝逆動き補償（フレームＡからフレームＢを予測するために計算される動きベクトルは、フレームＢからフレームＡを予測するために逆に使用される）。

According to the following notation. A = reference frame, B = current frame, DWT = discrete wavelet transform. A _d = low frequency spatial subband of DWT of frame A, ie low spatial resolution version of frame A. B _d = Low frequency spatial subband of frame B DWT, ie, low spatial resolution version of frame B. H = high frequency temporal subband with low spatial resolution. L = low frequency temporal subband at low spatial resolution. MC _down = motion compensation performed on reduced resolution (ie subsampled) frames. MC ⁻¹ = inverse motion compensation (the motion vector calculated to predict frame B from frame A is used inversely to predict frame A from frame B).

Equations (6)-(9) make it possible to define L _S and H _S.

ここで、Ｘ_S＝（Ｘ_S＝Ｈ_S又はＬ_Sによる）所与のフレームＸのＤＷＴの３つの高周波空間サブバンドの結合体。ＭＣ_full＝フル解像度のフレームに実行される動き補償。Ｌ’及びＨ’＝従来の３Ｄサブバンド方式における低周波時間サブバンド及び高周波時間サブバンド。Ｈ＝ＤＷＴ^-1［Ｈ_d∪Ｈ_S］。Ｌ＝ＤＷＴ^-1［Ｌ_d∪Ｌ_S］。

Where X _S = the combination of the three high frequency spatial subbands of the DWT of a given frame X (by X _S = H _S or L _S ). MC _full = motion compensation performed on full resolution frames. L ′ and H ′ = low frequency time subband and high frequency time subband in the conventional 3D subband system. H = DWT ⁻¹ [H _d ∪H _S ]. L = DWT ⁻¹ [L _d ∪L _S ].

  Once all of the low frequency time subbands and the high frequency time subband are generated at a given time level jt, the low frequency time subband L is the next time level jt + 1 at both low and high spatial resolution. Is further decomposed to achieve

  This ultimately results in a time-resolved structure that is very similar to the structure of the classic 3D subband encoder at each step of time-resolved. The last level low frequency time subbands and all levels of high frequency time subbands are then spatially decomposed through a wavelet filter and encoded to form a bitstream.

  The described invention retains good behavior for previous solutions at lower resolutions while approaching the performance of classic 3D subband codecs at full resolution (the overall resolution tree in 3D subband analysis) And no extra information is sent out to compensate for drift effects (disassembly / reconstruction mechanism is changed). Major upgrades arise from a new approach to generate high frequency spatial subbands that result in high consistency to the decomposition tree, thus improving the coding efficiency of the system.

At the decoder, all previous equations can be reverted to allow good reconstruction. Only ^ is added to each subband to indicate that decoding is involved and some information may be lost. First, classical 3D subband synthesis at low resolution makes it possible to give low spatial resolution subbands A _d and B _d from L _d and H _d .

また、Ｈを合成すること、及び式（７）を復帰させることでＡ_Sを得ることは容易である。プロセスは、式（１２）〜式（１５）により説明される。

Moreover, it is easy to obtain A _S by synthesizing H and returning Equation (7). The process is described by equations (12)-(15).

次いで、

Then

は、 (Outside 1)

Is

から簡単に再構成される。結果的に、Ｂ_Sを得ることができ、最終的にＢを合成することができる。このことは、式（１６）〜式（１９）のシステムにより要約される。 (Outside 2)

Easily reconstructed from. As a result, B _S can be obtained, and finally B can be synthesized. This is summarized by the system of equations (16)-(19).

これらの演算は、文字通り最初の時間レベルまで、すなわちＧＯＦが完全に復号化されるまで繰り返される。Ｌ及びＨがビットストリームで完全に送信されるとすぐ、完全な再構成が達成されるので、この方式はドリフトを生じないことが明らかに分かるであろう（フル空間解像度の合成がそれぞれの時間レベルで低い解像度のものと即座にリンクされる。これは、先のソリューションにおけるケースではない）。

These operations are literally repeated until the first time level, that is, until the GOF is fully decoded. As soon as L and H are completely transmitted in the bitstream, it will clearly be seen that this scheme does not cause drift, since full reconstruction is achieved (combining full spatial resolution at each time). Immediately linked to the lower resolution level, which is not the case in the previous solution).

  The previously defined coding principle illustrates the main process of the coding method in FIG. 6 and the corresponding motion compensated temporal filtering scheme in a more detailed manner (actually two diagrams, FIG. 7A). And with reference to FIG. 7 (including FIG. 7B).

  In the coding scheme of FIG. 6, the original group of frames GOF (this current GOF includes a full resolution frame FRF) is used to generate a low resolution frame LRF by wavelet decomposition WD. The motion compensated space-time analysis MCSTA is then performed on this low resolution frame. In this way, a low resolution sequence is obtained. Also, since the original full resolution frame (ie, each full resolution GOF) produces a high spatial subband HSS, motion compensated spatial-temporal analysis (corresponding successive steps MCSTA and WD are "MC temporal analysis" and Used to perform “corresponding to“ wavelet decomposition ”).

After a set of these two parallel steps performed on a full resolution frame, the low frequency subbands thus obtained for the decomposition are low resolution sequences LRS according to the following behavior, respectively, at the time resolution level: Are repeatedly replaced by their corresponding space-time subbands.
(A) First, the storage operation 62 stores the high-frequency space-time subbands related to the decomposition in consideration of the last encoding step 69.
(B) The wavelet synthesis 63 is then performed from the low frequency space-time subband of the decomposition (test 61 “L or H time subband” Allows separation of space-time subbands).
(C) The test 64 for the rank of the time resolution level then stores the low frequency space-time subband for the decomposition if the level is the last, and if the level is not the last In contrast, two parallel sets of steps are further performed for the next time level 66.

A more detailed representation of the overall decomposition scheme (on the encoding side) and the corresponding motion compensation synthesis scheme (on the decoding side) comprises FIG. 7 and (two diagrams, FIGS. 8A and 8B). It can be seen in FIG. This example of space-time decomposition according to the invention relates to a GOF consisting of four frames A0 to A3 (for clarity) with forward motion compensation and two decomposition levels. High and low frequency (respectively H ′ ₀ , H ′ ₁ and L ′ ₀ , L ′ ₁ ) time subbands are described, for example, in the literature “Factoring wavelet transforms into lifting steps”, I. Daubechies and W. Sweldens, Bell It is calculated from the original frame using the so-called lifting method described in Laboratories technical report, Lucent Technologies, 1996. The notations DWT and DWT ^-1 indicate wavelet decomposition and wavelet synthesis, respectively. The right side of FIG. 7 shows the inverse synthesis applied to the low frequency space-time subband for the first space-time resolution level decomposition and the low frequency sub for the decomposition by the corresponding space-time subband of the low resolution sequence. It is performed after the band replacement, which continuously illustrates the second space-time resolution level (indicated by the arrow from the left side of FIG. 7).

  Although the video encoding method and apparatus according to the present invention has been described above in a detailed manner, the present invention is dual in terms of the steps performed when implementing the video encoding method. The invention relates to a corresponding video decoding method comprising successive steps and to a corresponding video decoding device comprising means which are continuous with the means provided in the video coding apparatus.

FIG. 3 shows 3D subband decomposition. It is a figure which shows embodiment regarding the encoding system which concerns on previous embodiment. It is a figure which illustrates motion compensation time analysis in the lowest resolution. It is a figure which shows embodiment regarding the decoding system corresponding to the encoding system of FIG. FIG. 6 is a diagram illustrating reordering (rearrangement) of high spatial subbands (for forward motion compensation). It is a figure which illustrates the main processes of the encoding method which concerns on this invention. It is a figure which illustrates the decomposition | disassembly method of corresponding motion compensation time filtering. It is a figure which illustrates the decomposition | disassembly method of corresponding motion compensation time filtering. FIG. 6 is a diagram illustrating the realization of a synthesis scheme corresponding to the encoding method of FIG. 5 on the decoding side. FIG. 6 is a diagram illustrating the realization of a synthesis scheme corresponding to the encoding method of FIG. 5 on the decoding side.

Claims (4)

A video encoding method for compression of an original video sequence that is divided into successive groups of frames (GOF), comprising:
Generating a sequence of low resolution frames organized into a continuous low resolution GOF from the full resolution frames of the original video sequence by wavelet decomposition;
Performing a motion compensated space-time analysis on each low resolution GOF of the low resolution frame sequence to obtain a low resolution sequence;
Performing a motion compensated spatio-temporal analysis of each full resolution GOF of the original video sequence;
At each temporal resolution level, replacing the low frequency subbands for the decomposition with corresponding space-time subbands of the low resolution sequence;
Encoding the resulting modified sequences and motion vectors generated during motion compensated spatial-temporal analysis of each full resolution GOF to generate an output encoded bitstream;
A video encoding method comprising: A video encoding device for compression of an original video sequence that is divided into successive groups of frames (GOF),
Means for generating, from the full resolution frame of the original video sequence, a sequence of low resolution frames organized into a continuous low resolution GOF by wavelet decomposition;
Means for performing a motion compensated space-time analysis on each low resolution GOF of the low resolution frame sequence to obtain a low resolution sequence;
Means for performing motion compensation space-time analysis on each full resolution GOF of the original video sequence;
Means for replacing the low frequency subbands for the decomposition with corresponding space-time subbands of the low resolution sequence at each time resolution level;
Means for encoding the resulting modified sequences and motion vectors generated during motion compensated spatial-temporal analysis for each full resolution GOF to generate an output encoded bitstream;
A video encoding apparatus comprising: A video decoding method provided for decoding an encoded bitstream corresponding to a video sequence encoded by a video encoding method, comprising:
The video encoding method is used to compress the original video sequence.
Generating a sequence of low resolution frames organized into a continuous low resolution GOF from the full resolution frames of the original video sequence by wavelet decomposition;
Performing a motion compensated space-time analysis on each low resolution GOF of the low resolution frame sequence to obtain a low resolution sequence;
Performing a motion compensated space-time analysis on each full resolution GOF of the original video sequence;
At each temporal resolution level, replacing the low frequency subbands for the decomposition with corresponding space-time subbands of the low resolution sequence;
Encoding the resulting modified sequences and motion vectors generated during motion compensated spatial-temporal analysis for each full resolution GOF to generate an output encoded bitstream;
The video decoding method includes a process performed according to the video encoding method of claim 1 and a binary continuous process.
A video decoding method characterized by the above. A video decoding device provided for decoding an encoded bitstream corresponding to a video sequence encoded by a video encoding device, comprising:
The video encoding device is for compressing the original video sequence.
Means for generating, from the full resolution frame of the original video sequence, a sequence of low resolution frames organized into a continuous low resolution GOF by wavelet decomposition;
Means for performing a motion compensated space-time analysis on each low resolution GOF of the low resolution frame sequence to obtain a low resolution sequence;
Means for performing motion compensation space-time analysis on each full resolution GOF of the original video sequence;
Means for replacing the low frequency subbands for the decomposition with corresponding space-time subbands of the low resolution sequence at each time resolution level;
Means for encoding the resulting modified sequences and motion vectors generated during motion compensated spatial-temporal analysis of each full resolution GOF to generate an output encoded bitstream;
The video decoding apparatus includes means continuous with the means provided in the video encoding apparatus of claim 2 and binary continuous means,
A video decoding apparatus characterized by the above.

Images