|Title of Invention||
"METHOD FOR ENCODING AND DECODING AN INTERLACED SEQUENCE OF DIGITAL VIDEO DATA AND AN APPARATUS THEREOF"
|Abstract||A method for encoding an interlaced sequence of digital video data, wherein a base layer for basic reconstruction of the video data and one or more enhancement layers for enhancing the base layer reconstruction are generated, characterized in that; splitting (SP) the interlaced sequence (I) into two sub¬sequences (Fl, F2); - spatially filtering (SF1) the first sub-sequence (Fl), wherein a lower frequency stream (L) and a higher frequency stream (H) is generated; -spatio-temporally filtering (STF) the lower frequency stream (L), wherein a first base layer signal (P) for reconstruction of a video sequence in progressive node of first resolution is generated; - performing a first notion estimation (ME1) on said lower frequency stream (L), wherein the resulting motion vectors (MV1) together with said first base layer signal (P) form a second base layer signal (BL), for reconstruction of a video sequence in progressive mode of second resolution; - motion-compensated temporal filtering (MCTF) the second sub¬sequence (F2) to obtain enhancement layer signals (FF2).|
|Full Text|| The present invention relates to a method for encoding and decoding a interlaced sequence of digital video data and an apparatus thereof.
Field of the invention
This invention relates to a method for video compressing. More particularly, it relates to a method for achieving scalability when encoding and decoding an interlaced sequence of digital video data.
Today most video coding algorithms use motion compensated predictive coding for reduction of transmission bandwidth. In such hybrid schemes, temporal redundancy is reduced using motion compensation, and spatial redundancy is reduced by motion compensation residual transform coding. While most standard definition (SD) video material, including available MPEG video material for test purpose, is interlaced, recent research in scalable video coding has concentrated on evaluation of progressive video only. Efforts in this respect are encouraged e.g. for MPEG-4 as mentioned in "Description of Exploration experiments in Scalable Video Coding", ISO/IEC JEC1/SC29/WG11, MPEG2002/N5414, Awaji, December 2002. Known algorithms, e.g. as disclosed in WO9003082 for HDTV, are not sufficient for transmission of spatially and temporally scalable, interlaced video, because a wider range of scalability is demanded. New display types, e.g. miniature mobile displays, get by with progressive video sequences of low spatial and temporal resolution, while others, e.g. HDTV
receivers, require interlaced sequences with high spatial resolution and standard temporal resolution.
Summary_of_ the_ Invention
The purpose of the invention is an improvement of subband or wavelet coding of interlaced video. Since in interlaced video the spatial and temporal dimensions are tightly coupled, the inventive method uses additional exploration in the context of spatio-temporal signal-noise-ratio (SNR).
The problem t.o be solved by the invention is to provide a method for full spatial and temporal scalability in coding and decoding of interlaced video. Such method is disclosed for coding in claim 1 and for decoding in claim 6. An apparatus that utilizes the method for coding is disclosed in claim 10. An apparatus that utilizes the method for decoding is disclosed in claim 11. It can covert the received signal to video sequences at different levels of temporal resolution, either progressive or interlaced, and at various levels of spatial resolution.
According to the invention, interlaced digital video data are processed in a way that supports spatial, temporal and quality scalability, quality with respect to SNR, namely by adequate splitting operations and spatio-temporal filtering, leading to the generation of scalable spatial and/or temporal sub-resolution sequences.
This is achieved by partitioning, or decomposing, the interlaced source video material into two layers, namely a base layer (BL) and an enhancement layer (EL). The BL contains
a fully scalable description of the input video source in progressive mode, wherein scalability refers to spatial resolution, temporal resolution and quality. The EL, when combined with the BL, allows a low-cost, scalable reconstruction of the original interlaced video material, including the achievement of original spatial resolution, temporal resolution and SNR quality.
This decomposition may be achieved by associating every other field of the interlaced video sequence to the same layer, e.g. one layer processes even fields and the other layer processes odd fields. Advantageously, it is possible to save computing power by using motion estimation (ME) vectors from the BL for parts of EL processing. The interlaced video sequence is decomposed into five components, belonging to the two layers BL and EL. A receiver according to the invention can combine the components to video sequences of various quality levels, with respect to spatio-temporal resolution and SNR.
Advantageous embodiments of the invention are disclosed in the dependent claims, the following description and the figures.
Brief description of the drawings
Exemplary embodiments of the invention are described with reference to the accompanying drawings, which show in
Fig.l the structure of a video coder for scalable coding according to the invention; and
Fig.2 the structure of a scalable video decoder according to the invention.
Detailed descrirption__of_ the invention
Fig.l shows the structure of a video coder according to the invention. The input signal I is an interlaced video sequence with a temporal resolution of 2α Hz, i.e. 2α fields per second or a frames per second, and a spatial resolution of 2Mx2N, i.e. a frame having 2N rows with 2M pixels each, or a field having N rows with 2M pixels each. A splitting block SP distributes the sequence such that the fields are passed in an alternating manner to a first or a second branch. The first branch generates BL information and EL information, but will be referred to herein as BL branch. The other branch generates only EL information and will be referred to herein as EL branch. Each branch receives a video sequence with a temporal resolution of α frames per second and a spatial resolution of 2MxN pixels. E.g., the BL sequence Fl contains even fields and the EL sequence contains odd fields. Generally, the splitting block SP executes any splitting transformation, producing two descriptions from a single source description. It may also utilize various filters for more sophisticated processing, e.g. low/high filtering.
The BL branch performs spatial filtering SF1 on the Fl sequence along the x-axis, i.e. within lines, and generates from the lower frequency portions of Fl a video sequence L with a spatial resolution of MxN pixels, and from the higher frequency portions of Fl a spatial resolution enhancement signal H. In a simple example the filtering can be performed by a low pass filter LProw and a high pass filter HProw , both performing horizontal subsampling of e.g. factor 2. Other factors may also appear useful. The low-resolution sequence L is input, to a spatio-temporal filter STF and to a Motion
Estimating (ME) block ME1. The spatio-temporal filter STF performs motion compensated temporal transformation, also being called 3D wavelet transform or 2D+t wavelet transform. It may be implemented e.g. as Haar transform filter of length 2, or Daubechies filter of length 9 and 7, or length 5 and 3. The ME block ME1 generates motion vectors MV1, which serve three purposes: first they are used to optimize the spatio-temporal filtering process, second they are transmitted to the receiver to be used e.g. for the inverse filtering process, and third they will be used in the EL branch of the coder.
The output signal P of the spatio-temporal filter STF is transmitted, and can be combined with the motion vectors MV1 by a receiver to reproduce e.g. an MxN pixel, a Hz progressive mode video sequence. Advantageously, when using the above mentioned implementation it is also possible to design the spatio-temporal filter such that this progressive sequence is scalable with respect to temporal resolution and to spatial resolution in both x and y direction. The filter may contain several filtering levels, depending on the target frame rate or format, so that e.g. a spatial resolution of M′xN′ and/or a temporal resolution of θ α Hz can be achieved, with M' and N' ranging in [0,M] [0,N] and θ ε Q01, Q01 being the rational numbers ranging in [0,1]. This filtering can deliver different spatio-temporal frequencies whose fittings, e.g. organization within the data stream, make the base layer fully scalable. The output P of this block STF is a progressive sequence with a resolution of M′xN′, which is a sub-resolution of the original 2Mx2N format, and it represents the lowest spatial and temporal resolution level the inventive system generates.
The spatial resolution enhancement signal H, containing the high frequency fields along the x-axis, belongs to the
enhancement layer EL, and can be used by advanced receivers to reconstruct frames with higher spatial resolution in x direction, up to the original resolution.
The EL branch performs Motion Compensated Temporal Filtering (MCTF) on the F2 sequence. It uses motion vectors (MV) in order to minimize the amount of data to be transmitted when an object moves its position in one picture relative to another picture. Two different sets of MVs can be used: either MVs generated from the original interlaced sequence, or MVs generated by comparing fields of same parity, either odd or even. Advantageously, for the latter case the MV1 vectors generated in the ME1 block of the BL branch can be used, since it is very likely that the motion between e.g. two successive even fields is very similar to the motion between the two corresponding odd fields, or vice versa. Still it is possible that motion changes within a sequence of four fields, so that it may be advantageous to use better motion vectors MV2, generated from the interlaced sequence I in a second motion estimation block ME2. Therefore a decision block is integrated in the EL branch in order to decide which set of MVs is to be used for MCTF. It may be advantageous to transmit the decision information.
Motion compensated temporal filtering (MCTF) makes it possible to generate filtered F2 fields FF2, i.e. F2 fields with lower cost than the original F2 fields. These filtered fields FF2 have the same spatial and temporal resolution as the input F2 fields. Cost reduction is achieved by using the described motion vectors, since the amount of transmitted data is reduced. Motion estimation and motion compensation can be performed across fields, independent from parity, in order to capture intra- and inter-fields correlation. Various
prediction modes (backward, forward or bi-directional) as well as a multiple reference fields algorithm can be used. In order to reduce MV coding cost and the estimation complexity, motion information from the base layer may be used a priori or a posteriori, depending on the targeted requirements.
For transmission, it will be advantageous to multiplex the resulting streams into one bitstream. The described encoding method decomposes the interlaced original video sequence in two layers, BL and EL, which allows the resulting bitstream to support a variety of spatial and temporal resolution levels, at the coder side as well as at the decoder side. The described EL may also be viewed upon as a plurality of layers, since it contains several levels of resolution.
As a concrete example, the input video sequence I is standard definition (SD) video, with α = 30 or 2α Hz being 60 Hz, Fl represents e.g. the even fields and F2 e.g. the odd fields, LProw performs 2:1 subsampling, L is a Common Image Format (CIF) sequence at 30 Hz, and STF is a 1-level spatio-temporal filter, so that P is a CIF sequence at 15 Hz.
At: the decoder side, several scenarios have to be considered, depending on the targeted applications: lossless or lossy decoding in progressive or interlaced modes. Fig.2 shows an exemplary advanced decoder that can generate various quality levels, selectable by technical or user needs. The decoder mode can be selected by a quality selector QS. This can be controlled e.g. by the transmitter or content provider, or the technical capabilities of the receiver, or user preferences. Examples for different modes are lossless or lossy, different field rates, different spatial resolution etc. The decoding process depends on the selected decode quality.
The simplest mode, from the decoder point of view, is progressive lossless mode. In this case only the base layer signals BL, comprising a progressive sequence P and motion vectors MV1, need to be fully decoded. Decoding is performed using a spatio-temporal synthesis STS. In this manner, the reconstructed video sequence V1 has the full spatio-temporal resolutions, i.e. spatial resolution of MxN pixel and temporal resolution of α Hz.
Another mode is the progressive lossy mode. This means that the spatio-temporal resolution of the reconstructed video sequence V2 is a sub-resolution of the resolution achieved in progressive lossless mode V1. The usage of spatio-temporal synthesis STS for decoding makes it possible to reconstruct any M'xN', θ'α Hz progressive sequence, with M' and N' ranging in ]0,M] and ]0,N] and θ' ranging in ]0,1]. Again, only the base layer needs to be decoded.
When the full original interlaced video material shall be reconstructed, i.e. without loss of information, interlaced lossless mode is required. This means that both the entire base layer BL and the enhancement layer EL must be fully decoded. As described for the coder, the EL signals comprise additional data H related to the higher frequencies in x direction of Fl fields, motion vectors MV2 generated from the input interlaced sequence, and information FF2 related to the F2 fields. For proper decoding all the coder operations have to be performed inversely. Inverse filtering is commonly called synthesis, When the respective inverse filtering, which is in this case spatial synthesis SS, is applied to the additional data H and the reconstructed BL signal V1, it produces a video sequence V3, which corresponds to the
original Fl sequence at. the coder side, of resolution 2MxN and a Hz. Further, the original F2 sequence from the coder side, with same resolution, can be synthesized from the other EL signals FF2 aNd MV2, by using an inverse motion compensated temporal filter IMCTF. The Fl fields are then merged by an inverse splitting operation ISP with the F2 fields in order to reconstruct the complete interlaced video sequence V4, with the full spatio-temporal resolutions of 2Mx2N and 2α Hz, or a interlaced frames per second.
As another possibility, interlaced lossy mode can be used. This means that the reconstructed spatio-temporal resolution is a sub-resolution of the full resolution. As described above for progressive lossy mode, decoding makes it possible to reconstruct any M'xN', θα Hz interlaced sequence. This lossy case means that the base layer or the enhancement layer, or both, are not completely decoded. This may be performed by the spatio-temporal synthesis STS, so that V2 can be used instead of V1 for further processing. It may also be performed by the spatial synthesis SS or the inverse motion compensated tempora1 fi1tering IMCTF.
In another embodiment of the invention it is possible to simplify the spatio-temporal filter STF at the coder so that the Fl sequence is directly transmitted. In this case it is also possible to cancel the spatio-temporal synthesis STS at the receiver.
Advantageously the inventive method provides a possibility to generate scalable video, including interlaced video, with the same bitstream being usable for a high resolution interlaced mode TV and a low resolution progressive mode mobile device. Since the bitstream is fully embedded, it is possible to
extract all desired spatial or temporal sub-resolutions, at the transmitter as well as at the receiver side.
The inventive method can be used for all video applications, preferably all MPEG compliant systems.
1. A method for encoding an interlaced sequence of digital video data, wherein a base layer for basic reconstruction of the video data and one or more enhancement layers for enhancing the base layer reconstruction are generated, characterized in that;
splitting (SP) the interlaced sequence (I) into two sub-sequences (Fl, F2);
spatially filtering (SF1) the first sub-sequence (Fl), wherein a lower frequency stream (L) and a higher frequency stream (H) is generated;
spatio-temporally filtering (STF) the lower frequency stream (L), wherein a first base layer signal (P) for reconstruction of a video sequence in progressive node of first resolution is generated;
performing a first notion estimation (ME1) on said lower frequency stream (L), wherein the resulting motion vectors (MV1) together with said first base layer signal (P) form a second base layer signal (BL), for reconstruction of a video sequence in progressive mode of second resolution;
motion-compensated temporal filtering (MCTF) the second sub-sequence (F2) to obtain enhancement layer signals (FF2).
2. Method as claimed in claim 1, wherein said spatio-temporal filtering (STF) modifies the video resolution in temporal or spatial, or both, dimensions.
3. Method as claimed in claim 1 or 2, comprising the step of performing a second motion estimation (ME2) on the interlaced video sequence (I), wherein second motion vectors (MV2) are generated.
4. Method as claimed in any of claims 1-3, wherein said motion-compensated digital filtering (MCTF) uses the interlaced video signal (I) and/or results of the first motion estimation (ME1).
5. Method as claimed in any of claims 1-4, wherein at least some of the following sub-sequences are transmitted: the first base layer signal (P), the motion vectors (MV1) generated from the first sub-sequence (Fl), the higher frequency stream (H) generated from the first subsequence (Fl), the motion vectors (MV2) generated from the interlaced sequence (I), the enhancement layer signals (FF2) obtained from the motion-compensated temporal filtering (MCTF).
6. A method for decoding a plurality of substreams of a bitstream representing an interlaced sequence of video data, wherein the single substreams are associated with either a base layer (BL) or an enhancement layer (EL), characterized in that:
extracting from the bitstream those substreams (P, MV1) that are associated with the base layer (BL); and
performing spatio-temporal synthesis (STS) on said sequence (VI) is generated;
selecting or determining (QS) a decode quality being either a basic or an enhanced decode quality, wherein control signals from a transmitter or content provider, technical capabilities at the receiver side and/or user preferences are evaluated;
upon selection of an enhanced decode quality, performing spatial synthesis (SS) on said progressive mode video sequence (VI) and a substream (H) associated with the enhancement layer (EL), wherein the spatial synthesis (SS) generates an enhanced progressive mode video sequence (V3).
7. Method as claimed in claim 6, comprising the steps
performing an inverse motion-compensated temporal filtering (IMCTF) on said enhanced progressive mode video sequence (V3) and one or more of the substreams (MV2, FF2) being associated with the enhancement layer (EL);
performing inverse splitting (ISP) on the output of said inverse motion-compensated temporal filtering and said enhanced progressive node video sequence (V3) wherein an interlaced video sequence (V4) of enhanced or full spatio-temporal resolution is generated.
8. Method as claimed in claim 6 or 7, wherein said spatio-
temporal synthesis (STS) on the substreams associated with the base
layer (BL) produces a scaled representation of a progressive mode video
sequence (V2), the scaling referring to spatial, temporal or both
9. An apparatus for performing a method as claimed in claim 1,
characterized in that
a splitting block (SP) for decomposing the interlaced sequence (I) into two sub-sequences (F1,F2);
a spatial filter (SF1) for decomposing the first sub-sequence (Fl) into a lower frequency stream (L) and a higher frequency stream (H);
a spatio-temporal filter (STF) for processing the lower frequency stream (L), wherein a first base layer signal (P) for reconstruction of a video sequence in progressive mode of a first resolution is generated;
a first motion estimation block (ME1) for processing said lower frequency stream (L), wherein the resulting motion vectors (MV1) together with said first base layer signal (P) form a second base layer signal (BL) for reconstruction of a video sequence in progressive mode of a second resolution; and
a motion-compensated temporal filter (MCTF) for processing the second sub-sequence (F2) to obtain enhancement layer signals (FF2).
10. An apparatus for performing a method as claimed in claim 6
characterized in that;
spatio-temporal synthesis block (STS) for generating from substreams (P, MV1) associated with the base layer a progressive mode video sequence (VI) of a first resolution;
a quality selector block (QS) for selecting or determining a decode quality being either a basic or an enhanced decode quality, wherein control signals from a transmitter or content provider, technical capabilities at the receiver side and/or user preferences are evaluated;
a spatial synthesis block (SS) for processing, upon selection of an enhanced decode quality in said quality selector block (QS), said progressive mode video sequence (VI) and a substream (H) associated with the enhancement layer (EL) the spatial synthesis block (SS) generating an enhanced progressive mode video sequence (V3).
11. Apparatus as claimed in claim 10, comprising
an inverse motion-compensated temporal filtering block (IMCTF) for processing said enhanced progressive mode video sequence (V3) and one or more of the substreams (MV2, FF2) being associated with the enhancement layer(EL);
an inverse splitting block (ISP) for processing the output of said inverse motion-compensated temporal filtering block (IMCTF) and said enhanced progressive mode video sequence (V3), wherein an interlaced video sequence (V4) of enhanced or full spatio-temporal resolution is generated.
|Indian Patent Application Number||3540/DELNP/2005|
|PG Journal Number||32/2008|
|Date of Filing||09-Aug-2005|
|Name of Patentee||THOMSON LICENSING|
|Applicant Address||46, QUAI A. LE GALLO, 92100 BOULOGNE-BILLANCOURT, FRANCE.|
|PCT International Classification Number||H04N 7/26|
|PCT International Application Number||PCT/EP2004/001658|
|PCT International Filing date||2004-02-20|