Patent Application
20190215513
The motivation for increased coding efficiency in video coding has led to the adoption in the Joint Video Team (JVT) (a standards body) of more refined and complicated models and modes describing motion information for a given macroblock. These models and modes tend to make better advantage of the temporal redundancies that may exist within a video sequence. See, for example, ITU-T, Video Coding Expert Group (VCEG), “JVT Coding—(ITU-T H.26L & ISO/IEC JTC1 Standard)—Working Draft Number 2 (WD-2)”, ITU-T JVT-B118, March 2002; and/or Heiko Schwarz and Thomas Wiegand, “Tree-structured macroblock partition”, Doc. VCEG-N17, December 2001.
The recent models include, for example, multi-frame indexing of the motion vectors, increased sub-pixel accuracy, multi-referencing, and tree structured macroblock and motion assignment, according to which different sub areas of a macroblock are assigned to different motion information. Unfortunately these models tend to also significantly increase the required percentage of bits for the encoding of motion information within sequence. Thus, in some cases the models tend to reduce the efficacy of such coding methods.
Even though, in some cases, motion vectors are differentially encoded versus a spatial predictor, or even skipped in the case of zero motion while having no residue image to transmit, this does not appear to be sufficient for improved efficiency.
It would, therefore, be advantageous to further reduce the bits required for the encoding of motion information, and thus of the entire sequence, while at the same time not significantly affecting quality.
Another problem that is also introduced by the adoption of such models and modes is that of determining the best mode among all possible choices, for example, given a goal bitrate, encoding/quantization parameters, etc.
Currently, this problem can be partially solved by the use of cost measures/penalties depending on the mode and/or the quantization to be used, or even by employing Rate Distortion Optimization techniques with the goal of minimizing a Lagrangian function.
Such problems and others become even more significant, however, in the case of Bidirectionally Predictive (B) frames where a macroblock may be predicted from both future and past frames. This essentially means that an even larger percentage of bits may be required for the encoding of motion vectors.
Hence, there is a need for improved method and apparatuses for use in coding (e.g., encoding and/or decoding) video data.
This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.
In view of the above, improved video coding is described to encode video data within a sequence of video frames. To this end, at least a portion of a reference frame is encoded to include motion information associated with the portion of the reference frame. At least a portion of a predictable frame that includes video data predictively correlated to said portion of said reference frame is defined based on the motion information. At least said portion of the predictable frame is encoded without including corresponding motion information and including mode identifying data. The mode identifying data indicate that the encoded portion of the predictable frame can be directly derived using at least the motion information associated with the portion of the reference frame
The present invention is illustrated by way of example and not limitation in the figures of the accompanying drawings. The same numbers are used throughout the figures to reference like components and/or features.
In accordance with certain aspects of the present invention, methods and apparatuses are provided for coding (e.g., encoding and/or decoding) video data. The methods and apparatuses can be configured to enhance the coding efficiency of “interlace” or progressive video coding streaming technologies. In certain implementations, for example, with regard to the current H.26L standard, so called “P-frames” have been significantly enhanced by introducing several additional macroblock Modes. In some cases it may now be necessary to transmit up to 16 motion vectors per macroblock. Certain aspects of the present invention provide a way of encoding these motion vectors. For example, as described below, Direct P prediction techniques can be used to select the motion vectors of collocated pixels in the previous frame.
While these and other exemplary methods and apparatuses are described, it should be kept in mind that the techniques of the present invention are not limited to the examples described and shown in the accompanying drawings, but are also clearly adaptable to other similar existing and future video coding schemes, etc.
Before introducing such exemplary methods and apparatuses, an introduction is provided in the following section for suitable exemplary operating environments, for example, in the form of a computing device and other types of devices/appliances.
Turning to the drawings, wherein like reference numerals refer to like elements, the invention is illustrated as being implemented in a suitable computing environment. Although not required, the invention will be described in the general context of computer-executable instructions, such as program modules, being executed by a personal computer.
Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. Those skilled in the art will appreciate that the invention may be practiced with other computer system configurations, including hand-held devices, multi-processor systems, microprocessor based or programmable consumer electronics, network PCs, minicomputers, mainframe computers, portable communication devices, and the like.
The invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.
The improved methods and systems herein are operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of well known computing systems, environments, and/or configurations that may be suitable include, but are not limited to, personal computers, server computers, thin clients, thick clients, hand-held or laptop devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like.
As shown in
Bus 136 represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. By way of example, and not limitation, such architectures include Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, Enhanced ISA (EISA) bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnects (PCI) bus also known as Mezzanine bus.
Computer 130 typically includes a variety of computer readable media. Such media may be any available media that is accessible by computer 130, and it includes both volatile and non-volatile media, removable and non-removable media.
In
Computer 130 may further include other removable/non-removable, volatile/non-volatile computer storage media. For example,
The drives and associated computer-readable media provide nonvolatile storage of computer readable instructions, data structures, program modules, and other data for computer 130. Although the exemplary environment described herein employs a hard disk, a removable magnetic disk 148 and a removable optical disk 152, it should be appreciated by those skilled in the art that other types of computer readable media which can store data that is accessible by a computer, such as magnetic cassettes, flash memory cards, digital video disks, random access memories (RAMs), read only memories (ROM), and the like, may also be used in the exemplary operating environment.
A number of program modules may be stored on the hard disk, magnetic disk 148, optical disk 152, ROM 138, or RAM 140, including, e.g., an operating system 158, one or more application programs 160, other program modules 162, and program data 164.
The improved methods and systems described herein may be implemented within operating system 158, one or more application programs 160, other program modules 162, and/or program data 164.
A user may provide commands and information into computer 130 through input devices such as keyboard 166 and pointing device 168 (such as a “mouse”). Other input devices (not shown) may include a microphone, joystick, game pad, satellite dish, serial port, scanner, camera, etc. These and other input devices are connected to the processing unit 132 through a user input interface 170 that is coupled to bus 136, but may be connected by other interface and bus structures, such as a parallel port, game port, or a universal serial bus (USB).
A monitor 172 or other type of display device is also connected to bus 136 via an interface, such as a video adapter 174. In addition to monitor 172, personal computers typically include other peripheral output devices (not shown), such as speakers and printers, which may be connected through output peripheral interface 175.
Computer 130 may operate in a networked environment using logical connections to one or more remote computers, such as a remote computer 182. Remote computer 182 may include many or all of the elements and features described herein relative to computer 130.
Logical connections shown in
When used in a LAN networking environment, computer 130 is connected to LAN 177 via network interface or adapter 186. When used in a WAN networking environment, the computer typically includes a modem 178 or other means for establishing communications over WAN 179. Modem 178, which may be internal or external, may be connected to system bus 136 via the user input interface 170 or other appropriate mechanism.
Depicted in
In a networked environment, program modules depicted relative to computer 130, or portions thereof, may be stored in a remote memory storage device. Thus, e.g., as depicted in
Attention is now drawn to
With the examples of
This section presents a new highly efficient Inter Macroblock type that can significantly improve coding efficiency especially for high/complex motion sequences. This Inter Macroblock new type takes advantage of the temporal and spatial correlations that may exist within frames at the macroblock level, and as a result can significantly reduce the bits required for encoding motion information while retaining or even improving quality.
Direct Prediction
The above mentioned problems and/or others are at least partially solved herein by the introduction of a “Direct Prediction Mode” wherein, instead of encoding the actual motion information, both forward and/or backward motion vectors are derived directly from the motion vectors used in the correlated macroblock of the subsequent reference frame.
This is illustrated, for example, in
Introduced herein, in accordance with certain implementations of the present invention, is a new Inter Macroblock type is provided that can effectively exploit spatial and temporal correlations that may exist at the macroblock level and in particular with regard to the motion vector information of the macroblock. According to this new mode it is possible that a current macroblock may have motion that can be directly derived from previously decoded information (e.g., Motion Projection). Thus, as illustratively shown in
Similar to the Direct Mode in B frames, by again assuming that speed is constant, motion for a macroblock can be directly derived from the correlated macroblock of the reference frame. This is further illustrated in
It is even possible to consider acceleration for refining such motion parameters, for example, see
The process may also be significantly improved by, instead of considering motion projection at the macroblock level, taking into account that the pixels inside the previous image are possibly moving with a constant speed or a constant acceleration (e.g., Pixel Projection). As such, one may generate a significantly more accurate prediction of the current frame for B frame coding as illustrated, for example, in
In certain implementations it is also possible to combine both methods together for even better performance.
In accordance with certain further implementations, motion can also be derived from spatial information, for example, using prediction techniques employed for the coding of motion vectors from the motion information of the surrounding macroblocks. Additionally, performance can also be further enhanced by combining these two different methods in a multi-hypothesis prediction architecture that does not require motion information to be transmitted. Consequently, such new macroblock types can achieve significant bitrate reductions while achieving similar or improved quality.
Exemplary Encoding Processes:
Video data 1004 is provided to a summation module 1006, which also receives as an input, the output from a motion compensation (MC) module 1022. The output from summation module 1006 is provided to a discrete cosine transform (DCT) module 1010. The output of DCT module 1010 is provided as an input to a quantization module (QP) 1012. The output of QP module 1012 is provided as an input to an inverse quantization module (QP−1)1014 and as an input to a variable length coding (VLC) module 1016. VLC module 1016 also receives as in input, an output from a motion estimation (ME) module 1008. The output of VLC module 1016 is an encoded video bitstream 1210.
The output of QP−1 module 1014 is provided as in input to in inverse discrete cosine transform (DCT) module 1018. The output of 1018 is provided as in input to a summation module 1020, which has as another input, the output from MC module 1022. The output from summation module 1020 is provided as an input to a loop filter module 1024. The output from loop filter module 1024 is provided as an input to a frame buffer module 1026. One output from frame buffer module 1026 is provided as an input to ME module 1008, and another output is provided as an input to MC module 1022. Me module 1008 also receives as an input video data 1004. An output from ME 1008 is proved as an input to MC module 1022.
In this example, MC module 1022 receives inputs from ME module 1008. Here, ME is performed on a current frame against a reference frame. ME can be performed using various block sizes and search ranges, after which a “best” parameter, using some predefined criterion for example, is encoded and transmitted (INTER coding). The residue information is also coded after performing DCT and QP. It is also possible that in some cases that the performance of ME does not produce a satisfactory result, and thus a macroblock, or even a subblock, could be INTRA encoded.
Considering that motion information could be quite costly, the encoding process can be modified as in
Motion can be modeled, for example, in any of the following models or their combinations: (1) Motion Projection (e.g., as illustrated in
Other prediction models (e.g. acceleration, filtering, etc.) may also be used. If only one of these models is to be used, then this should be common in both the encoder and the decoder. Otherwise, one may use submodes which will immediately guide the decoder as to which model it should use. Those skilled in the art will also recognize that multi-referencing a block or macroblock is also possible using any combination of the above models.
In
Here, video encoder 1202 has been modified to include improvement 1204. Improvement 1204 includes an additional motion vector (MV) buffer module 1206 and a DIRECT decision module 1208. More specifically, as shown, MV buffer module 1206 is configured to receive as inputs, the output from frame buffer module 1026 and the output from ME module 1008. The output from MV buffer module 1206 is provided, along with the output from ME module 1008, as an input to DIRECT decision module 1208. The output from DIRECT decision module 1208 is then provided as an input to MC module 1022 along with the output from frame buffer module 1026.
For the exemplary architecture to work successfully, the Motion Information from the previously coded frame is stored intact, which is the purpose for adding MV buffer module 1206. MV buffer module 1206 can be used to store motion vectors. In certain implementations. MV buffer module 1206 may also store information about the reference frame used and of the Motion Mode used. In the case of acceleration, for example, additional buffering may be useful for storing motion information of the 2nd or even N previous frames when, for example, a more complicated model for acceleration is employed.
If a macroblock, subblock, or pixel is not associated with a Motion Vector (i.e., a macroblock is intra coded), then for such block it is assumed that the Motion Vector used is (0, 0) and that only the previous frame was used as reference.
If multi-frame referencing is used, one may select to use the motion information as is, and/or to interpolate the motion information with reference to the previous coded frame. This is essentially up to the design, but also in practice it appears that, especially for the case of (0, 0) motion vectors, it is less likely that the current block is still being referenced from a much older frame.
One may combine Direct Prediction with an additional set of Motion Information which is, unlike before, encoded as part of the Direct Prediction. In such a case the prediction can, for example, be a multi-hypothesis prediction of both the Direct Prediction and the Motion Information.
Since there are several possible Direct Prediction submodes that one may combine, such could also be combined within a multi-hypothesis framework. For example, the prediction from motion projection could be combined with that of pixel projection and/or spatial MV prediction.
Direct Prediction can also be used at the subblock level within a macroblock. This is already done for B frames inside the current H.26L codec, but is currently only using Motion Projection and not Pixel Projection or their combinations.
For B frame coding, one may perform Direct Prediction from only one direction (forward or backward) and not always necessarily from both sides. One may also use Direct Prediction inside the Bidirectional mode of B frames, where one of the predictions is using Direct Prediction.
In the case of Multi-hypothesis images, for example, it is possible that a P frame is referencing to a future frame. Here, proper scaling, and/or inversion of the motion information can be performed similar to B frame motion interpolation.
Run-length coding, for example, can also be used according to which, if subsequent “equivalent” Direct P modes are used in coding a frame or slice, then these can be encoded using a run-length representation.
DIRECT decision module 1208 essentially performs the decision whether the Direct Prediction mode should be used instead of the pre-existing Inter or Intra modes. By way of example, the decision may be based on joint Rate/Distortion Optimization criteria, and/or also separate bitrate or distortion requirements or restrictions.
It is also possible, in alternate implementations, that module Direct Prediction module 1208 precedes the ME module 1008. In such case, if Direct Prediction can provide immediately with a good enough estimate, based on some predefined conditions, for the motion parameters, ME module 1008 could be completely by-passed, thus also considerably reducing the computation of the encoding.
Exemplary Decoding Processes:
Reference is now made to
Encoded video data bitstream 1104 is provided as an input to a variable length decoding (VLD) module 1106. The output of VLD module 1106 is provided as an input to a QP−1 module 1108, and as an input to an MC module 1110. The output from QP−1 module 1108 is provided as an input to an IDCT module 1112. The output of IDCT module 1112 is provided as an input to a summation module 1114, which also receives as an input an output from MC module 1110. The output from summation module 1114 is provided as an input to a loop filter module 1116. The output of loop filter module 1116 is provided to a frame buffer module 1118. An output from frame buffer module 1118 is provided as an input to MC module 1110. Frame buffer module 1118 also outputs (decoded) video data 1120.
An exemplary improved decoder 1302 for use in a Direct Prediction environment 1300 further includes an improvement 1306. Here, as shown in
Improvement 1306, in this example, is operatively inserted between MC module 1110 and a VLD module 1106′. Improvement 1306 includes an MV buffer module 1308 that receives as an input, an output from VLD module 1106′. The output of MV buffer module 1308 is provided as a selectable input to a selection module 1312 of improvement 1306. A block mode module 1310 is also provided in improvement 1306. Block mode module 1310 receives as an input, an output from VLD module 1106′. An output of block mode module 1310 is provided as an input to VLD module 1106′, and also as a controlling input to selection module 1312. An output from VLD module 1106′ is provided as a selectable input to selection module 1312. Selection module 1312 is configured to selectably provide either an output from MV buffer module 1308 or VLD module 1106′ as an input to MC module 1110.
With improvement 1306, for example, motion information for each pixel can be stored, and if the mode of a macroblock is identified as the Direct Prediction mode, then the stored motion information, and the proper Projection or prediction method is selected and used. It should be noted that if Motion Projection is used only, then the changes in an existing decoder are very minor, and the additional complexity that is added on the decoder could be considered negligible.
If submodes are used, then improved decoder 1302 can, for example, be configured to perform steps opposite to the prediction steps that improved encoder 1202 performs, in order to properly decode the current macroblock.
Again non referenced pixels (such as intra blocks) may be considered as having zero motion for the motion storage.
Some Exemplary Schemes
Considering that there are several possible predictors that may be immediately used with Direct Prediction, for brevity purposes in this description a smaller subset of cases, which are not only rather efficient but also simple to implement, are described in greater detail. In particular, the following models are examined in greater demonstrative detail:
Wherein, residue is not transmitted for QP<n (e.g., n=24). Here, run-length coding is not used. The partitioning of the submodes can be set as follows:
The best submode could be selected using a Rate Distortion Optimization process (best compromise between bitrate and quality).
Motion Vector Prediction in Bidirectionally Predictive (B) Frames with Regards to Direct Mode:
The current JVT standard appears to be quite unclear on how a Direct Mode coded macroblock or block should be considered in the motion vector prediction within Bidirectionally Predicted (B) frames. Instead, it appears that the current software considers a Direct Mode Macroblock or subblock as having a “different reference frame” and thus not used in the prediction. Unfortunately, considering that there might still be high correlation between the motion vectors of a Direct predicted block with its neighbors such a condition could considerably hinder the performance of B frames and reduce their efficiency. This could also reduce the efficiency of error concealment algorithms when applied to B frames.
In this section, exemplary alternative approaches are presented, which can improve the coding efficiency increase the correlation of motion vectors within B frames, for example. This is done by considering a Direct Mode coded block essentially equivalent to a Bidirectionally predicted block within the motion prediction phase.
Direct Mode Macroblocks or blocks (for example, in the case of 8×8 sub-partitions) could considerably improve the efficacy of Bidirectionally Predicted (B) frames since they can effectively exploit temporal correlations of motion vector information of adjacent frames. The idea is essentially derived from temporal interpolation techniques where the assumption is made that if a block has moved from a position (x+dz, y+dy) at time t to a position (x,y) at time t+2, then, by using temporal interpolation, at time t+1 the same block must have essentially been at position:
This is illustrated, for example, in
The later is illustrated in
Unfortunately, the current JVT standard does not clarify how the motion vector prediction of blocks adjacent to Direct Mode blocks should be performed. As it appears from the current software, Direct Mode blocks are currently considered as having a “different reference frame” thus no spatial correlation is exploited in such a case. This could considerably reduce the efficiency of the prediction, but could also potentially affect the performance of error concealment algorithms applied on B frames in case such is needed.
By way of example, if one would like to predict the motion vector of E in the current codec, if A, B, C, and D were all Direct Mode coded, then the predictor will be set as (0,0) which would not be a good decision.
In
In accordance with certain aspects of the present invention, instead one may use the actual Motion information available from the Direct Mode coded blocks, for performing the motion vector prediction. This will enable a higher correlation of the motion vectors within a B frame sequence, and thus can lead to improved efficiency.
One possible issue is how to appropriately handle Direct Mode Macroblocks for which, the collocated block/macroblock in the subsequent frame was intra coded. Here, for example, two possible options include:
In accordance with certain other exemplary implementations of the present invention, a further modification can be made in the de-blocking filter process. For the Direct Mode case, a de-blocking filter process can be configured to compare stored motion vector information that is taken from Direct Mode coded blocks—otherwise these would usually be considered as zero. In another modification, however, instead one may configure the de-blocking filter process to compare the (exact) motion vectors regardless of the block type that is used. Thus, in certain implementations, if for Direct Coded blocks no residue is transmitted, a “stronger” de-blocking filter can provide further improved performance.
Furthermore, in certain other implementations, the Rate Distortion Decision for B frames can be redesigned since it is quite likely that for certain implementations of the motion vector prediction scheme, a different langrangian parameter A used in Rate Distortion Optimization decisions, may lead to further coding efficiency. Such A can be taken, for example, as:
Inter Mode Decision Refinement:
The JVT standard currently has an overwhelming performance advantage versus most other current Block Based coding standards. Part of this performance can be attributed in the possibility of using variable block sizes raging from 16×16 down to 4×4 (pixels), instead of having fixed block sizes. Doing so, for example, allows for a more effective exploitation of temporal correlation. Unfortunately, it has been found that, due to the Mode Decision techniques currently existing in conventional coding logic (e.g., hardware, firmware, and/or software), mode decisions might not be optimally performed, thus wasting bits that could be better allocated.
In this section, further methods and apparatuses are provided that at least partly solve this problem and/or others. Here, the exemplary methods and apparatuses have been configured for use with at least 16×8 and 8×16 (pixel) block modes. Furthermore, using a relatively simple solution where at least one additional criterion is introduced, a saving of between approximately 5% and 10% is provided in the complexity of the encoder.
Two key features of the JVT standard are variable macroblock mode selection and Rate Distortion Optimization. A 16×16 (pixel) macroblock can be coded using different partitioning modes for which motion information is also transmitted. The selection of the mode to be used can be performed in the Rate Distortion Optimization phase of the encoding where a joint decision of best possible quality at best possible bitrate is attempted. Unfortunately, since the assignments of the best possible motion information for each subpartition is done in an entirely different process of the encoding, it is possible in some cases, that a non 16×16 mode (e.g. 16×8 or 8×16 (pixel)) carries motion information that is equivalent to a 16×16 macroblock. Since the motion predictors used for each mode could also be different, it is quite possible in many cases that such 16×16 type motion information could be different from the one assigned to the 16×16 mode. Furthermore, under certain conditions, the Rate Distortion Optimization may in the end decide to use the non 16×16 macroblock type, even though it continues 16×16 motion information, without examining whether such could have been better if coded using a 16×16 mode.
Recognizing this, an exemplary system can be configured to determine when such a case occurs, such that improved performance may be achieved. In accordance with certain exemplary implementations of the present invention, two additional modes, e.g., referred to as P2to1 and P3to1, are made available within the Mode decision process/phase. The P2to1 and P3to1 modes are enabled when the motion information of a 16×8 and 8×16 subpartitioning, respectively, is equivalent to that of a 16×16 mode.
In certain implementations all motion vectors and reference frame assigned to each partition may be equal. As such, the equivalent mode can be enabled and examined during a rate distortion process/phase. Since the residue and distortion information will not likely change compared to the subpartition case, they can be reused without significantly increasing computation.
Considering though that the Rate Distortion Mode Decision is not perfect, it is possible that the addition and consideration of these two additional modes regardless of the current best mode may, in some limited cases, reduce the efficiency instead of improving it. As an alternative, one may enable these modes only when the corresponding subpartitioning mode was also the best possible one according to the Mode decision employed. Doing so may yield improvements (e.g., bitrate reduction) versus the other logic (e.g., codecs, etc.), while not affecting the PSNR.
If the motion information of the 16×8 or 8×16 subpartitioning is equivalent to that of the 16×16 mode, then performing mode decision for such a mode may be unnecessary. For example, if the motion vector predictor of the first subpartition is exactly the same as the motion vector predictor of the 16×16 mode performing mode decision is unnecessary. If such condition is satisfied, one may completely skip this mode during the Mode Decision process. Doing so can significantly reduce complexity since it would not be necessary, for this mode, to perform DCT, Quantization, and/or other like Rate Distortion processes/measurements, which tend to be rather costly during the encoding process.
In certain other exemplary implementations, the entire process can be further extended to a Tree-structured macroblock partition as well. See, e.g., Heiko Schwarz and Thomas Wiegand, “Tree-structured macroblock partition”, Doc. VCEG-N17, December 2001.
An Exemplary Algorithm
Below are certain acts that can be performed to provide a mode refinement in an exemplary codec or other like logic (note that in certain other implementations, the order of the act may be changed and/or that certain acts may be performed together):
Applying Exemplary Direct Prediction Techniques for Interlace Coding:
Due to the increased interest of interlaced video coding inside the H.26L standard, several proposals have been presented on enhancing the encoding performance of interlaced sequences. In this section techniques are presented that can be implemented in the current syntax of H.26L, and/or other like systems. These exemplary techniques can provide performance enhancement. Furthermore, Direct P Prediction technology is introduced, similar to Direct B Prediction, which can be applied in both interlaced and progressive video coding.
Further Information on Exemplary Direct P Prediction Techniques:
Direct Mode of motion vectors inside B-frames can significantly benefit encoding performance since it can considerably reduce the bits required for motion vector encoding, especially considering that up to two motion vectors have to be transmitted. If, though, a block is coded using Direct Mode, no motion vectors are necessary where as instead these are calculated as temporal interpolations of the motion vectors of the collocated blocks in the first subsequent reference image. A similar approach for P frames appears to have never been considered since the structure of P frames and of their corresponding macroblock was much simpler, while each macroblock required only one motion vector. Adding such a mode would have instead, most likely, incurred a significant overhead, thus possibly negating any possible gain.
In H.26L on the other hand, P frames were significantly enhanced by introducing several additional macroblock Modes. As described previously, in many cases it might even be necessary to transmit up to 16 motion vectors per macroblock. Considering this additional Mode Overhead that P frames in H.26L may contain, an implementation of Direct Prediction of the motion vectors could be viable. In such a way, all bits for the motion vectors and for the reference frame used can be saved at only the cost of the additional mode, for example, see
Even though a more straightforward method of Direct P prediction is to select the Motion vectors of the collocated pixels in the previous frame, in other implementations one may also consider Motion Acceleration as an alternative solution. This comes from the fact that maybe motion is changing frame by frame, it is not constant, and by using acceleration better results could be obtained, for example, see
Such techniques can be further applied to progressive video coding. Still, considering the correlation that fields may have in some cases inside interlace sequences, such as for example regions with constant horizontal only movement, this approach can also help improve coding efficiency for interlace sequence coding. This is in particular beneficial for known field type frames, for example, if it is assumed that the motion of adjacent fields is the same. In this type of arrangement, same parity fields can be considered as new frames and are sequentially coded without taking consideration of the interlace feature. Such is entirely left on the decoder. By using this exemplary Direct P mode though, one can use one set of motion vectors for the first to be coded field macroblock (e.g., of size 16×16 pixels) where as the second field at the same location is reusing the same motion information. The only other information necessary to be sent is the coded residue image. In other implementations, it is possible to further improve upon these techniques by considering correlations between the residue images of the two collocated field Blocks.
In order to allow Direct Mode in P frames, it is basically necessary to add one additional Inter Mode into the system. Thus, instead of having only 8 Inter Modes, in one example, one can now use 9 which are shown below:
In general, such exemplary Direct Modes for P frames can appear if the collocated macroblock was also of INTER type, except Skip macroblock, but including Direct Mode, since in other cases there is no motion information that could be used. In the case of the previous macroblock also being coded in Direct P Mode, the most recent Motion Vectors and Mode for this macroblock are considered instead. To more efficiently though handle the cases that this Mode will not logically appear, and in particular if INTRA mode was used, one may select of allowing this Mode to also appear in such cases with the Mode now signifying a second Skip Macroblock Mode where a copy the information is not from the previous frame, but from the one before it. In this case, no residue information is encoded. This is particularly useful for Interlace sequences, since it is more likely that a macroblock can be found with higher accuracy from the same parity field frame, and not from the previously coded field frame as was presented in previous techniques.
For further improved efficiency, if a set of two Field type frames is used when coding interlace images, the Skip Macroblock Mode can be configured to use the same parity field images. If Direct P mode is used as a skipping flag, for example, then the different parity is used instead. An additional benefit of Direct P mode, is that it may allow for a significant complexity reduction in the encoder since it is possible to allow the system to perform a pre-check to whether the Direct P mode gives a satisfactory enough solution, and if so, no additional computation may be necessary for the mode decision and motion estimation of that particular block. To also address the issue of motion vector coding, the motion vectors used for Direct P coding can be used “as is” for the calculation of a MEDIAN predictor.
Best Field First Technique & Field Reshuffling:
Coding of interlaced sequence allowing support of both interlace frame material, and separate interlace field images inside the same stream would likely provide a much better solution than coding using only one of the two methods. The separate interlace field technique has some additional benefits, such as, for example, de-blocking, and in particular can provide enhanced error resilience. If an error happens inside one field image, for example, the error can be easily consumed using the information from the second image.
This is not the case for the frame based technique, where especially when considering the often large size of and bits used by such frames, errors inside such a frame can happen with much higher probability. Reduced correlation between pixels/blocks may not promote error recovery.
Here, one can further improve on the field/frame coding concept by allowing the encoder to select which field should be encoded first, while disregarding which field is to be displayed first. This could be handled automatically on a decoder where a larger buffer will be needed for storing a future field frame before displaying it. For example, even though the top field precedes the bottom field in terms of time, the coding efficiency might be higher if the bottom field is coded and transmitted first, followed by the top field frame. The decision may be made, for example, in the Rate Distortion Optimization process/phase, where one first examines what will the performance be if the Odd field is coded first followed by the Even field, and of the performance if the Even field is instead coded and is used as a reference for the Odd field. Such a method implies that both the encoder and the decoder should know which field should be displayed first, and any reshuffling done seamlessly. It is also important that even though the Odd field was coded first, both encoder and decoder are aware of this change when indexing the frame for the purpose of INTER/INTRA prediction. Illustrative examples of such a prediction scheme, using 4 reference frames, are depicted in
In the case of coding joint field/frame images, the scheme illustratively depicted in
Although the description above uses language that is specific to structural features and/or methodological acts, it is to be understood that the invention defined in the appended claims is not limited to the specific features or acts described. Rather, the specific features and acts are disclosed as exemplary forms of implementing the invention.
This patent application is a continuation of U.S. patent application Ser. No. 10/186,284 (“parent”), titled “Improved Video Coding Methods and Apparatuses”, filed on Jun. 27, 2002, and hereby incorporated by reference. The parent claimed benefit priority from: (1) U.S. patent application Ser. No. 60/376,005, filed Apr. 26, 2002; and (2) U.S. patent application Ser. No. 60/352,127, filed Jan. 15, 2002.
| Number | Date | Country | |
|---|---|---|---|
| 60376005 | Apr 2002 | US | |
| 60352127 | Jan 2002 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11275103 | Dec 2005 | US |
| Child | 12474821 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 15867361 | Jan 2018 | US |
| Child | 16353914 | US | |
| Parent | 13850178 | Mar 2013 | US |
| Child | 15867361 | US | |
| Parent | 12474821 | May 2009 | US |
| Child | 13850178 | US | |
| Parent | 10186284 | Jun 2002 | US |
| Child | 11275103 | US |
