JP2006340014A - Low complexity motion compensated temporal direction filtering method and apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a motion compensation type time direction filter processing method capable of reducing complexity and delay caused by motion compensation type time direction filter processing, in consideration that a conventional motion compensation type time direction filter processing method imposes a motion picture decoding device to perform inverse update processing, and inverse prediction processing so that a great deal of complexity for the device and a large amount of delay in decoding processing are required. <P>SOLUTION: Inverse update processing is performed immediately after decoding processing of an inter-picture encoding macroblock in wide area pictures. Consequently, a delay time for restoration processing and complexity in a decoding device can be reduced. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention is used in any multimedia data coding, and in particular, can be used in video coding corresponding to motion compensation type time direction filtering.

  In recent years, the development of video compression technology has been developed in an effort to improve the coding efficiency that exceeds the conventional H.264 / AVC, and motion-compensated temporal filtering (motion-compensated temporal filtering) (MCTF)) seen in the expansion of use. The difference between video coding based on MCTF and conventional video coding such as ISO / IEC 11172-2, ISO / IEC 13818-2, and ISO / IEC14496-2 Part 10 is the prediction residual component. The linearly combined component is added to the pixel of the input signal by the update step, and a temporal low frequency component is obtained.

  The lifting method used in MCTF can be expressed as follows.

  Here, a signal represents a prediction / update operator, a high-frequency signal, and a low-frequency signal, respectively. There are two types of filters that are most often used in motion compensated temporal filtering. They are a Haar filter and a 5/3 conversion filter.

In the Haar type filter, is expressed as follows.

  In the 5/3 conversion filter, the prediction / update operator is obtained as follows.

In order to reconstruct pixels based on the lifting method, the reverse update process must be performed before the reverse prediction process. Therefore, the reconstruction process can be expressed as follows.

  Here, denotes a signal, denotes a prediction / update operator, denotes a reconstructed high-frequency signal, and denotes a reconstructed low-frequency signal.

  Currently, in a moving image codec such as ISO / IEC 14496-10 Amendment 2 Scalable Video Coding, both a Haar type filter and a 5/3 conversion filter can be used for motion compensation type time direction filter processing. Since the update processes in these two filters are different from each other, the moving picture decoding apparatus must first decode the high-frequency picture and store it in the memory of the high-frequency decoding apparatus. When sufficient high-frequency pictures are decoded and stored in the picture buffer of the decoding device, the decoding device then uses the high-frequency pictures to perform reverse update processing with reverse prediction processing on the low-frequency pictures. Would do against.

  FIG. 1 is a block diagram showing a configuration of a conventional picture coding apparatus. The picture coding apparatus includes a picture memory 500, a motion detection unit 502, a motion compensation unit 504, a difference calculation unit 506, an orthogonal transformation unit 508, a quantization unit 510, an entropy coding unit 512, a residual picture memory 514, an updated motion vector. A calculation unit 516, an update process selection unit 518, a second motion compensation unit 520, a switch 522, a weight update unit, and an addition unit 526 are provided.

A picture input to the picture coding apparatus is first stored in the picture memory 500. The motion detection unit 502 calculates motion vectors, ref_idx_10 and ref_idx_11 based on a comparison between Pict _Even and Pict _Odd from the picture memory. The motion compensation unit 504 uses the motion vector, ref_idx_10, and ref_idx_11 information to calculate a predicted picture from the picture memory. The difference calculation unit 506 subtracts the predicted picture from Pict _Even in order to output the residual of the high frequency picture. This residual is input to the orthogonal transform unit 508 and output as a transform coefficient. Next, the quantization unit 510 quantizes this coefficient and outputs a quantization coefficient QuantCoeff. The entropy encoding unit 512 encodes the QuantCoeff, the motion vector, ref_idx_10, and ref_idx_11 into one bit stream. The residual output by the difference calculation unit 506 is stored in the residual picture memory 514. The update motion vector calculation unit obtains the motion vectors, ref_idx_10 and ref_idx_11 output from the motion detection unit, and calculates the update motion vectors, update_ref_idx_10 and update_ref_idx_11. The update process selection unit 518 acquires the residual from the residual picture memory, calculates the energy intensity of the residual, and outputs a decision as to whether or not to perform the update process to the switch 522 and the weight to the weight update unit 524. The motion compensation unit 520 performs motion compensation on the residual picture in the residual picture memory 514 in order to output the updated residual UpdatedRes using the updated motion vectors, update_ref_idx_10 and update_ref_idx_11. The switch 522 is used to turn off the update process according to the determination output from the update process selection unit 518. The weight update unit 524 scales UpdatedRes by the weight output from the update process selection unit 518. Finally, the adding unit 526 updates Pict _Odd by adding the scaled Wi ^* UpdatedRes to Pict _odd , and stores the updated picture in the picture memory 500.

  FIG. 2 is a configuration diagram showing a configuration of a conventional picture decoding apparatus. The decoding apparatus includes an entropy decoding unit 400, an inverse quantization unit 402, an inverse orthogonal transform unit 404, a decoding residual memory 406, an update motion vector calculation unit 408, an update process selection unit 410, a motion compensation unit 412, and a switch 416. , A weight update unit 416, a difference calculation unit 418, a reconstructed picture memory 420, a second motion compensation unit 422, and an addition unit 424.

The entropy decoding unit 400 acquires an encoded input signal, and outputs quantized coefficients, motion vectors, ref_idx_10, and ref_idx_11 of the encoded signal. The inverse quantization unit 402 inversely quantizes the encoded signal into decoded coefficients. Next, the inverse orthogonal transform unit 404 acquires a decoding coefficient and converts it into a decoding residual DecodedRes. These decoded residuals are stored in the decoded residual memory 406. The update motion vector calculation unit 408 calculates the update motion vector, update_ref_idx_10, and update_ref_idx_11 using the motion vectors, ref_idx_10 and ref_idx_11 output from the entropy decoding unit 400. The update process selection unit 410 obtains the residual from the residual picture memory, calculates the energy intensity of the residual, and outputs a decision as to whether or not to perform the update process to the switch 414 and the weight to the weight update unit 416. The motion compensation unit 412 performs motion compensation on the residual picture in the decoded residual memory 406 in order to output the updated residual UpdatedRes using the updated motion vectors, update_ref_idx_10 and update_ref_idx_11. The switch 414 is used to turn off the update process according to the determination output from the update process selection unit 410. The weight update unit 416 scales UpdatedRes by the weight output from the update process selection unit 410. The difference calculation unit 418 subtracts the scaled residual Wi ^* UpdatedRes from the reconstructed picture in the reconstructed picture memory 420 and stores the updated picture in the reconstructed picture memory 420. The motion compensation unit 422 uses the motion vectors, ref_idx_10, and ref_idx_11 in order to predict from the reconstructed picture memory, and outputs a predicted picture Pred. The adding unit 424 calculates the sum of the prediction picture and the decoding residual in the decoding residual memory 406, and outputs the decoded picture. If the decoded picture is a reference picture, the decoded picture is also stored in the reconstructed picture memory 420.

  FIG. 3 is a diagram for explaining the concept of MCTF. The encoding process in MCTF is described as follows. In temporal direction level 4, odd pictures 1101, 1103, 1105, etc. are subtracted by predicted pictures from adjacent even pictures in order to obtain high-frequency pictures 1117, 1118, 1119, etc. These high frequency pictures are encoded into one bit stream. The even pictures 1100, 1102, 1104, 1106, etc. are then updated using predictions from the residuals of the high pass pictures to form low pass pictures 1125, 1126, 1127, etc. In order to generate a new high-frequency and low-frequency picture pair in the temporal direction level 2, prediction and update processing in the next step is performed on these low-frequency pictures. The low-pass pictures in the respective prediction and update processes are subjected to prediction and update processes in the next process until one low-pass picture can be used for GOP or partition. High-frequency pictures in all processes of prediction and update processing are stored in a bit stream. The low-frequency picture is encoded as an intra-picture encoded picture, or is encoded as a predicted picture predicted from a previously encoded picture belonging to another GOP.

  In the MCTF decoding process, an update process is first performed before the prediction process. The high frequency picture 1145 decoded from the bit stream is updated to the low frequency picture 1146 to be a low frequency picture 1144. The predicted picture from low pass picture 1144 is then added to high pass picture 1145 to generate low pass picture 1143. These update and prediction processing steps are repeated until all pictures in the time direction level 4 are reconstructed. The middle low-pass picture is reconstructed from the high-pass picture, but all the high-pass pictures are encoded into one bit stream.

  Both the inverse update process and the inverse prediction process in the prior art involve motion compensation with a 1/4 pixel accuracy. The problem is that both of them involve motion compensation with 1/4 pixel accuracy, so that the reverse update process becomes as complicated as the reverse prediction process. Furthermore, since the inverse update process can be performed only after the reconstruction of all corresponding high-frequency pictures, the complexity of the video decoding device is limited, resulting in a large delay in the time required for the reconstruction of pictures. This time delay is reduced in some decoding device implementations, but increases the cost of increasing the complexity of the decoding device, thus indirectly increasing the cost of the decoding device.

  In order to solve the above complexity problem in inverse update processing, a new method for low complexity motion compensated temporal direction filtering is proposed. The new point of the present invention is that, in the present invention, the moving picture decoding apparatus can perform the reverse update process immediately after the decoding process of the inter-picture encoded macroblock in the high-frequency picture. This reduces the time delay required for picture restoration and the complexity of the decoding device.

  The effect of the present invention is to reduce the complexity required to perform the update process in the motion compensation type time direction filter process, and to indirectly reduce the cost for the encoding apparatus and the decoding apparatus.

  In order to reduce the delay in the decoding process and the complexity of the decoding apparatus, the present invention uses a new 2-tap filter in place of the Haar filter used in the prior art shown in equations (3) and (4). The new 2-tap filter is shown in Equation 9 and Equation 10. The present invention also achieves compatibility between the update operators of two filters by using a modified inverse update operator in the 5/3 transform filter.

  In the new 2-tap filter, this is expressed as:

  In the 5/3 conversion filter, the prediction and update operators are obtained as follows.

  From Equations 10 and 12, it is clear from the above modification that the 2-tap filter update operator forms part of the 5/3 transform filter update operator. Therefore, it can be seen that the reverse update processing can be performed on each high-frequency picture regardless of the type of filter used. This eliminates the need to wait for all high-frequency pictures to be processed before the reverse update process. In the present invention, W1 can be defined as a 1/4 value or a 1/2 value.

  FIG. 4 is a block diagram showing the configuration of the picture coding apparatus according to the present invention. The picture coding apparatus includes a picture memory 700, a motion detection unit 702, a motion compensation unit 704, a difference calculation unit 706, an orthogonal transformation unit 708, a quantization unit 710, an entropy coding unit 712, an update processing selection unit 714, and an update weight selection. Unit 716, three switches 718, 720 and 726, two scaling units 722 and 724, an updated motion vector calculation unit 728, a motion update unit, and an addition unit 732.

A picture input to the picture coding apparatus is first stored in the picture memory 700. The motion detection unit 702 calculates motion vectors, ref_idx_10 and ref_idx_11 based on a comparison between Pict _Even and Pict _{Odd in} the picture memory. The motion compensation unit 704 calculates a prediction picture from the picture memory using the motion vector, ref_idx_10, and ref_idx_11 information. The difference calculation unit 706 subtracts the predicted picture from Pict _Even in order to output the residual of the high frequency picture. These residuals are input to the orthogonal transform unit 708 and output as transform coefficients. The quantization unit 710 next quantizes these coefficients and outputs a quantization coefficient QuantCoeff. The entropy encoding unit 712 encodes the QuantCoeff, the motion vector, ref_idx_10, and ref_idx_11 into one bit stream. The update weight selection unit 716 selects a weight used for the update process, and outputs UpdateWeightType information to the switches 720 and 726. This UpdateWeightType information is transmitted to the entropy encoding unit 712 and encoded into one bit stream.

  The switch 718 is controlled by the update process selection unit 714. The update process selection unit 714 uses the ref_idx_10 and ref_idx_11 information and outputs a determination as to whether or not the update process can be performed. When the above processing is possible, the residual output from the difference calculation unit 706 is scaled by the scaling unit 722 or 724 and input to the addition unit 732. The updated motion vector calculation unit 728 acquires a motion vector from the motion detection unit 702 and outputs an updated motion vector. Next, the motion update unit 730 acquires the outputs of the ref_idx_10 and ref_idx_11 and the updated motion vector calculation unit 728 from the motion prediction unit 702, and outputs LowPassBlk from the picture memory 700. The adder 732 adds the values of LowPassBlk and ScaleRes, and stores the sum in the picture memory 700.

  In another embodiment, the update weight selection unit 716, the switches 722 and 724, and the second scaling unit 724 may be omitted. In this case, UpdateWeightType is not encoded in the bitstream. The residual is scaled with a weight of W1 by one scaling unit. The value of W1 is 1/4 or 1/2.

  FIG. 5 is a block diagram showing a configuration of a picture decoding apparatus according to the present invention. The picture decoding apparatus includes an entropy decoding unit 600, an inverse quantization unit 602, an inverse orthogonal transform unit 604, an addition unit 606, an update processing selection unit 606, three switches 608, 610 and 616, two scaling units 612 and 614, A difference calculation unit 622, an updated motion vector calculation unit 618, a motion update unit 620, a decoded picture memory 624, and a motion compensation unit 626 are provided.

  The entropy decoding unit 600 acquires an encoded input signal, and outputs quantized coefficients, motion vectors, ref_idx_10, and ref_idx_11 of the encoded signal. The inverse quantization unit 602 performs inverse quantization on the encoded signal into a decoded coefficient. Next, the inverse orthogonal transform unit 604 obtains a decoding coefficient and converts it into a decoding residual DecodedRes. The update process selection unit 606 acquires ref_idx_10 and ref_idx_11, and controls whether or not the switch 610 performs the update process. The switches 610 and 616 are controlled by the UpdateWeightType decoded by the entropy decoding unit 600. UpdateWeightType is used to select one of two different scaling units 612 and 614 that are used to scale the decoding residual from the output of switch 608.

  The updated motion vector calculation unit 618 obtains a motion vector from the entropy decoding unit 600 and calculates an updated motion vector. The motion update unit 620 acquires update motion vectors, ref_idx_10 and ref_idx_11, and acquires a reconstructed block from the decoded picture memory. The difference calculation unit 622 subtracts the value of the scaled residual ScaledRes from the value of the reconstructed block ReconBlk, and stores the result of the subtraction back into the decoded picture memory 624. The motion compensation unit 626 outputs a predicted picture from the decoded picture memory using the motion vector, ref_idx_10, and ref_idx_11 information. The adding unit 606 adds the prediction picture and the decoding residual from the inverse orthogonal transform unit 604, and outputs a reconstructed picture. When this reconstructed picture is used to generate a predicted picture of another picture, the reconstructed picture is stored in the decoded picture memory 624.

  In another embodiment, the two switches 610 and 616 and the second scaling unit 614 may be omitted. UpdateWeightType is not output from the entropy decoding unit, and the residual is scaled by the weight of W1 by one scaling unit. The value of W1 is 1/4 or 1/2.

FIG. 6A is a diagram illustrating a stream structure of an encoded bit stream when the picture encoding apparatus of the present invention switches the scaling weight of the update scaling unit in units of pictures. FIG. 6B is a diagram illustrating a stream structure of an encoded bit stream when the picture encoding apparatus of the present invention switches the scaling weight of the update scaling unit in units of slices. Update weight type information UpdateWeightType that identifies one of the update scaling weights is included. The picture decoding apparatus of the present invention that decodes an encoded bitstream is the same type as the update scaling weight used for encoding by checking the update scaling weight type information encoded in the bitstream. Select the weight of.

  When the weight of update scaling is switched in units of pictures, UpdateWeightType information can be stored in the header 801 as shown in FIG. 6A. The header 801 includes information regarding parameters and one or more pictures. When the weight of update scaling is switched in units of slices, UpdateWeightType information can be stored in the slice header 802 as shown in FIG. 6B. The slice header 802 includes information on parameters and slice data accompanying the slice header.

  FIG. 7 shows the motion compensation type time direction filter processing in the picture coding apparatus of the present invention. As illustrated, in the module 900, motion detection is performed by comparing adjacent pictures in the picture memory 902 with the current picture. Next, in module 904, motion vectors and reference indices for prediction are determined. Then, in module 906, motion compensation is performed to generate a predicted picture by predicting from a picture in the picture memory using the motion vector and the reference index. In module 908, the macroblock residual is calculated. The motion vector, reference index, and residual are then encoded into a single bitstream at module 910. The above process is included in the prediction process of the motion compensation type time direction filter process.

  In module 918, the residual block to be processed is analyzed to determine if it will be used in the update process. As shown in module 920, when the processing target block is used for update processing, an update motion vector is determined in module 922. Based on the updated motion vector, the residual is updated to the reference frame previously used for motion compensation, as shown in module 928. The updated reference frame is stored in the picture memory.

  FIG. 8 shows inverse motion compensation type time direction filter processing in the picture decoding apparatus of the present invention. In module 100, a reference picture used for motion prediction is determined. This is done by decoding information in the encoded macroblock header that indicates which reference picture is used by the macroblock for prediction. Next, in module 102, the motion vector is decoded from the macroblock header. These motion vectors are used in the subsequent inverse prediction process. In module 104, the block residual is decoded from the video bitstream. In module 106, the decoded block residual is stored in a residual memory so that the residual can be used for inverse prediction processing. In the module 110, when the processing target block is used for the reverse update process, the processing target block is determined. As shown in module 112, if the block residual is updated, an updated motion vector is determined in module 114 and the block residual is updated in module 116 to the corresponding position in the reference frame indicated by the updated motion vector. The These updated motion vectors have an integer pixel accuracy. The process described above is included in the inverse update process of the inverse motion compensation type time direction filter process.

  In module 120, motion compensation is performed using a prediction motion vector and a reference index for prediction from an updated reference picture. Next, in module 122, the picture is reconstructed by adding the block residual of the picture memory to the prediction block derived by the motion compensation process. Then, when the reconstructed picture is used for prediction of another picture, this reconstructed picture is stored in the reference frame memory 128 as shown in the module 126.

  FIG. 9 shows a process of updating the block residual to the reference picture. First, in the module 200, a block of predicted pixels is obtained from a reference picture located in the picture memory 202 using an integer pixel update motion vector, as indicated by the update motion vector. Next, in module 204, the update scaling weight type UpdateWeightType is decoded from the bitstream. As shown in module 206, if UpdateWeightType is 1, the update residual is scaled by a factor of 1/4 in module 208. As shown in module 210, if UpdateWeightType is 0, the update residual is scaled by a factor of 1/2. Next, in module 212, the predicted pixels are added to the residual scaled for the picture encoder. In the case of a picture decoding device, in module 212, the scaled residual is subtracted from the predicted pixel. Then, in module 213, the subtraction sum is clipped to an unsigned 8-bit value. The reconstructed picture is stored back into the picture memory 216 as indicated by the updated motion vector in module 214.

  FIG. 10 shows a process for determining an updated motion vector from a predicted motion vector. First, in the module 300, the updated motion vector is set to have the same value as the predicted motion vector. Next, in module 302, if the residual block includes luminance pixels, the residual block is determined. If the processing target block is a luminance block, the accuracy of the luminance motion vector is determined in the module 306. For example, in ISO / IEC 14496-10 MPEG-4 AVC, the accuracy of the luminance motion vector is 1/4 pixel. If the block to be processed is not a luminance block, the accuracy of the luminance motion vector is determined in the module 304. When the encoding format is 4: 2: 0, the luminance motion vector is 1/8 pixel accuracy. When the encoding format is 4: 2: 2, the horizontal component of the luminance motion vector has 1/8 pixel accuracy, but the vertical component of the luminance motion vector has 1/4 pixel accuracy.

  Next, in the module 308, an offset value is set. The offset value varies depending on the accuracy of the motion vector. When the motion vector is 1/8 pixel precision, the offset value is set to 4. When the motion vector is 1/4 pixel accuracy, the offset value is set to 2. Next, in module 309, a shift coefficient is set. When the motion vector is 1/8 pixel precision, the shift coefficient is set to 3. When the motion vector is 1/4 pixel accuracy, the shift coefficient is set to 2. Then, in module 310, the offset value is added to the updated motion vector. In module 312, the summed sum is shifted right by the number of bits indicated by the shift factor. Thereafter, in module 314, the updated motion vector is shifted left by the same number of bits as indicated by the shift factor.

  FIG. 11 shows a process for performing the update process selection process in the present invention. First, in the module 1000, the macroblock type MBType is determined from the macroblock header. As shown in the module 1002, when MBType is Intra, the update process is skipped. If MBType is not Intra, module 1004 determines a reference index and ref_idx_10 from the bitstream. As shown in the module 1006, when ref_idx_10 is not 0, the update process to the reference frame referenced by ref_idx_10 is skipped. As shown in module 1006, if ref_idx_10 is 0, in module 1008, the update process is selected for the reference picture referenced by ref_idx_10.

  Next, in the module 1010, when the macroblock type is BiPred, the macroblock type is determined. If MBType is not BiPred, the update process selection process ends. When MBType is BiPred, the module 1012 determines the reference index ref_idx_11 from the bitstream. As shown in the module 1014, when ref_idx_11 is not 0, the update process for the reference frame referenced by ref_idx_11 is skipped. As shown in module 1014, if ref_idx_11 is 0, in module 1016, an update process is selected for the reference picture referenced by ref_idx_11.

FIG. 1 is a block diagram showing the configuration of a conventional picture coding apparatus. FIG. 2 is a block diagram showing a configuration of a conventional picture decoding apparatus. FIG. 3 is a diagram for explaining high-frequency pictures and low-frequency pictures in MCTF processing. FIG. 4 is a block diagram showing the configuration of the picture coding apparatus according to the present invention. FIG. 5 is a block diagram showing a configuration of a picture decoding apparatus according to the present invention. FIG. 6A is a block diagram showing a stream structure of an encoded bit stream output from the picture encoding apparatus of the present invention. FIG. 6B is a diagram illustrating a stream structure of an encoded bit stream output when the picture encoding apparatus of the present invention switches update weights in units of slices. FIG. 7 is a flowchart showing motion compensation type time direction filter processing in the picture coding apparatus of the present invention. FIG. 8 is a flowchart showing motion compensation time direction filter processing in the picture decoding apparatus of the present invention. FIG. 9 is a flowchart showing a process for performing the update process in the present invention. FIG. 10 is a flowchart showing a process for deriving an updated motion vector in the present invention. FIG. 11 is a flowchart showing processing for selecting update processing in the present invention.

Claims (41)

A low-complexity motion-compensated temporal filter method in picture encoding processing,
Perform motion detection (900),
Determining a motion vector and a reference picture (904);
Perform motion compensation (906),
Calculating a macroblock residual (908);
Encoding the macroblock residual, motion vector, and reference index into one bitstream (910);
Determining whether the processing block of the residual is used for update processing (918);
When the processing target block is used for update processing, an update motion vector is determined (922),
Update the residual block to the reference picture (928)
Including steps. Updating the block residual to the reference picture;
Obtaining predicted pixels from the reconstructed reference picture (200);
Update weight type for update is determined (204),
Check whether the UpdateWeightType is 1 (206),
If the UpdateWeightType is 1, the block residual is scaled by a factor of 1/4 (208),
If the UpdateWeightType is not 0 (comment: I think it is a mistake of “1”, but it can be left as it is wrong from the original), the block residual is scaled by a factor of 1/2 (210) ,
Adding the predicted pixel to the scaled residual (212);
Clip the calculation result to an 8-bit positive value (213);
Copy the clipped value back to the reference picture (214)
The method according to claim 1, further comprising: steps. A low-complexity motion compensated temporal direction filtering method in picture decoding processing, comprising:
Determine a reference picture (100);
Decoding the predicted motion vector (102);
Decoding the block residual (104);
Storing the block residual in a residual memory (106);
Determining whether the block is used for update processing (110);
If the block is updated, determine an updated motion vector (114);
If the block is updated, update the block residual to the reference picture (116);
Performing motion compensation based on the predicted motion vector (120);
Reconstructing the picture by adding the prediction value from the motion compensation to the residual (122);
If the picture is referenced by another picture, store the reconstructed picture in a reference frame memory (126)
Including steps. The method of claim 3, wherein the reference picture is determined by decoding information from a macroblock header. The low-complexity motion compensated temporal direction filter processing method according to claim 3, wherein the motion vector predictor is decoded from a macroblock header and used for inverse prediction processing. The process of determining whether the block is used for update processing is as follows:
Determine the type of macroblock (1000)
Check whether the macroblock type is Intra (1002),
If the type of the macroblock is Intra, return to the decision to skip the update process for the block,
If the macroblock type is not Intra, a variable ref_idx_10 is determined from the bitstream (1004),
Check whether the variable ref_idx_10 is 0 (1006),
If the variable ref_idx_10 is not 0, return to the decision to skip the update process for the reference frame referenced by the ref_idx_10,
If the variable ref_idx_10 is 0, the process returns to the decision to perform update processing on the reference frame referenced by the ref_idx_10 (1008),
Determining whether the macroblock type is BiPred (1010);
If the macroblock type is not BiPred, the process returns to the decision to skip the update process for the second reference frame of the block,
If the type of the macroblock is BiPred, a variable ref_idx_11 is determined from the bitstream (1012),
Check whether the variable ref_idx_11 is 0 (1014),
If the variable ref_idx_11 is not 0, return to the decision to skip the update process for the reference frame referenced by the ref_idx_11,
If the variable ref_idx_11 is 0, the process returns to the decision to perform update processing on the reference frame referenced by the ref_idx_11 (1016).
4. The low complexity motion compensated temporal direction filtering method according to claim 1 and 3, further comprising steps. The low-complexity motion compensated temporal direction filter processing method according to claim 1, 3 or 6, wherein the variables ref_idx10 and ref_idx_11 are used for specifying a reference picture in a reference picture memory. The process of determining the updated motion vector includes:
Setting the updated motion vector equal to the decoded predicted motion vector (300);
Determining whether the block is a luminance block (300);
If the block is a luminance block, determine the accuracy of the luminance motion vector (306);
If the block is not a luminance block, determine the accuracy of the chrominance motion vector (304);
Based on the accuracy of the predicted motion vector, an offset value is set (308),
A shift coefficient is set based on the accuracy of the predicted motion vector (309),
Adding the offset value to the value of the updated motion vector (310);
Shifting the value of the updated motion vector to the right by the number of bits indicated by the shift factor (312);
The value of the updated motion vector is shifted to the left by the number of bits indicated by the shift coefficient (314).
4. The low complexity motion compensated temporal direction filtering method according to claim 1 and 3, further comprising steps. The low-complexity motion compensation type time direction filter processing method according to claim 1, wherein the accuracy of the luminance motion vector is ¼ pixel. The low-complexity motion-compensated temporal filter according to claim 1, 3 or 8, wherein when the encoded video format is 4: 2: 0, the accuracy of the color difference motion vector is 1/8 pixel. Processing method. When the encoded moving image format is 4: 2: 2, the accuracy of the vertical component of the color difference motion vector is 1/4 pixel, and the accuracy of the horizontal component of the color difference motion vector is 1/8. 9. The low complexity motion compensated temporal direction filtering method according to claim 1, 3 and 8. The low-complexity motion-compensated temporal filter according to claim 1, 3 and 8, wherein when the encoded video format is 4: 4: 4, the accuracy of the color difference motion vector is 1/4 pixel. Processing method. The low-complexity motion-compensated temporal direction filtering method according to claim 1, 3 or 8, wherein when the accuracy of the motion vector is 1/4 pixel, the offset value is set to a value of 2. . The low-complexity motion-compensated temporal direction filtering method according to claim 1, 3 or 8, wherein when the accuracy of the motion vector is 1/8 pixel, the offset value is set to a value of 4. . 9. The low-complexity motion-compensated temporal direction filtering method according to claim 1, 3 and 8, wherein when the accuracy of the motion vector is 1/4 pixel, the shift coefficient is set to a value of 2. . 9. The low-complexity motion compensated temporal direction filtering method according to claim 1, 3 and 8, wherein when the accuracy of the motion vector is 1/8 pixel, the shift coefficient is set to a value of 3. . Updating the block residual to the reference picture;
Obtaining a predicted pixel from the reconstructed reference picture (200);
A weight type UpdateWeightType for the update process is determined (204);
Check whether the UpdateWeightType is 1 (206),
If the UpdateWeightType is 1, the block residual is scaled by a factor of 1/4 (208),
If the UpdateWeightType is not 0, the block residual is scaled by a factor of 1/2 (210),
Subtracting the scaled residual from the predicted pixel (212);
Clip the calculation result to an 8-bit positive value (213);
Copy the clipped value back to the reference picture (214)
The method according to claim 3, further comprising a step. The low-complexity motion-compensated temporal direction filtering method according to claim 1, 2, 3, or 17, wherein the prediction pixel is acquired from the reference picture using the updated motion vector. Clipping the calculation result to an 8-bit positive value includes:
If the calculation result is greater than 255, set the calculation result to a value of 255;
18. The low-complexity motion-compensated temporal direction filtering method according to claim 1, further comprising a step of setting the calculation result to a value of 0 when the pixel has a value smaller than 0. . Copying the clipped value back into the reference picture;
Using the updated motion vector to find an update position in the reference picture;
18. The low-complexity motion compensated temporal direction filtering method according to claim 1, further comprising: replacing a value of a pixel at the update position of the reference picture with the clipped value. . A low-complexity motion-compensated temporal filter device for picture encoding processing,
Means for performing motion detection (900);
Means for determining a motion vector and a reference picture (904);
Means for performing motion compensation (906);
Means for calculating a macroblock residual (908);
Means for encoding said macroblock residual, motion vector and reference index into one bitstream (910);
Means for determining whether the residual processing block is used for update processing (918);
When the block to be processed is used for update processing, means for determining an update motion vector (922);
Means (928) for updating the processing block of the residual with the reference picture
It is characterized by providing. The means for updating the processing block of the residual to the reference picture,
Means for obtaining a prediction pixel from the reconstructed reference picture (200);
Means for determining a weight type UpdateWeightType for update processing (204);
Means for confirming whether the UpdateWeightType is 1 (206);
When the UpdateWeightType is 1, means for scaling the block residual by a factor of 1/4 (208);
Means (210) for scaling the block residual by a factor of 1/2 if the UpdateWeightType is not 0;
Means for adding the predicted pixel to the scaled residual (212);
Means for clipping the calculation result to an 8-bit positive value (213);
Means for copying the clipped value back to the reference picture;
The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21, comprising: A low-complexity motion-compensated temporal filter device for picture coding processing,
Means for determining a reference picture (100);
Means for decoding the predicted motion vector (102);
Means for decoding the block residual (104);
Means for storing the block residual in a residual memory (106);
Means for determining whether said block is used for update processing (110);
Means for determining an updated motion vector if the block is updated (114);
Means for updating the block residual to the reference picture if the block is updated (116);
Means for performing motion compensation based on the predicted motion vector (120);
Means for reconstructing the picture by adding a prediction value from the motion compensation to the residual (122);
Means (126) for storing the reconstructed picture in a reference frame memory when the picture is referenced by another picture;
It is characterized by providing. The low-complexity motion compensated temporal filter apparatus according to claim 23, wherein the reference picture is determined by decoding information from a macroblock header. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 23, wherein the motion vector predictor is decoded from a macroblock header and used for inverse prediction processing. In the process of determining whether the block is used for update processing,
Means for determining the type of macroblock (1000);
Means for confirming whether the type of the macroblock is Intra (1002);
If the type of the macroblock is Intra, means for returning to the decision to skip the update process for the block;
If the type of the macroblock is not Intra, means for determining a variable ref_idx_10 from the bitstream (1004);
Means for checking whether the variable ref_idx_10 is 0 (1006);
If the variable ref_idx_10 is not 0, means for returning to the decision to skip the update process for the reference frame referenced by the ref_idx_10;
If the variable ref_idx_10 is 0, means for returning to the decision to perform update processing on the reference frame referenced by the ref_idx_10 (1008);
Means for determining whether the macroblock type is BiPred (1010);
Means for returning to the decision to skip the update process for the second reference frame of the block if the type of the macroblock is not BiPred;
When the type of the macroblock is BiPred, means for determining a variable ref_idx_11 from the bitstream (1012),
Means for checking whether the variable ref_idx_11 is 0 (1014);
If the variable ref_idx_11 is not 0, means for returning to the decision to skip the update process for the reference frame referenced by the ref_idx_11;
When the variable ref_idx_11 is 0, a means for returning to the decision to perform update processing on the reference frame referenced by the ref_idx_11 (1016)
24. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21 and 23. 27. The low-complexity motion-compensated temporal direction filter processing apparatus according to claim 21, 23, and 26, wherein the variables ref_idx10 and ref_idx_11 are used for specifying a reference picture in a reference picture memory. The means for determining the updated motion vector comprises:
Means (300) for setting the updated motion vector equal to the decoded predicted motion vector;
Means (300) for determining whether the block is a luminance block;
Means (306) for determining the accuracy of the luminance motion vector, if the block is a luminance block;
Means (304) for determining the accuracy of the chrominance motion vector if the block is not a luminance block;
Means (308) for setting an offset value based on the accuracy of the predicted motion vector;
Means (309) for setting a shift coefficient based on the accuracy of the predicted motion vector;
Means (310) for adding the offset value to the value of the updated motion vector;
Means (312) for shifting the value of the updated motion vector to the right by the number of bits indicated by the shift factor;
Means (314) for shifting the value of the updated motion vector to the left by the number of bits indicated by the shift coefficient
24. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21 and 23. 29. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21, 23, or 28, wherein the accuracy of the luminance motion vector is 1/4 pixel. 29. The low-complexity motion-compensated temporal filter according to claim 21, 23 and 28, wherein when the encoded video format is 4: 2: 0, the accuracy of the color difference motion vector is 1/8 pixel. Processing equipment. When the encoded moving image format is 4: 2: 2, the accuracy of the vertical component of the color difference motion vector is 1/4 pixel, and the accuracy of the horizontal component of the color difference motion vector is 1/8. 29. The low-complexity motion-compensated time-direction filter processing apparatus according to claim 21, 23 and 28. 29. The low-complexity motion-compensated temporal filter according to claim 21, 23 and 28, wherein when the encoded video format is 4: 4: 4, the accuracy of the color difference motion vector is 1/4 pixel. Processing equipment. 29. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21, 23, or 28, wherein when the accuracy of the motion vector is 1/4 pixel, the offset value is set to a value of 2. . 29. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21, 23, or 28, wherein when the accuracy of the motion vector is 1/8 pixel, the offset value is set to a value of 4. . 29. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21, 23, or 28, wherein when the accuracy of the motion vector is 1/4 pixel, the shift coefficient is set to a value of 2. . 29. The low-complexity motion compensated temporal direction filtering method according to claim 21, 23 and 28, wherein when the accuracy of the motion vector is 1/8 pixel, the shift coefficient is set to a value of 3. . Obtaining a predicted pixel from the reconstructed reference picture (200);
A weight type UpdateWeightType for the update process is determined (204);
Check whether the UpdateWeightType is 1 (206),
If the UpdateWeightType is 1, the block residual is scaled by a factor of 1/4 (208),
If the UpdateWeightType is not 0, the block residual is scaled by a factor of 1/2 (210),
Subtracting the scaled residual from the predicted pixel (212);
Clip the calculation result to an 8-bit positive value (213);
Copy the clipped value back to the reference picture (214)
24. The low complexity motion compensated temporal direction filter processing apparatus according to claim 23, wherein the block residual including the step is updated to the reference picture. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21, 22, 23, or 27, wherein the prediction pixel is acquired from the reference picture using the updated motion vector. The means for clipping the calculation result to an 8-bit positive value is:
Means for setting the calculation result to a value of 255 if the calculation result is a value greater than 255;
28. The low-complexity motion compensated temporal direction filter processing apparatus according to claim 21, comprising means for setting the calculation result to a value of 0 when the pixel has a value smaller than 0. . Means for copying the clipped value back to the reference picture;
Means for finding an update position in the reference picture using the update motion vector;
The low-complexity motion compensated temporal direction filter processing method according to claim 21, comprising means for replacing a value of a pixel at the update position of the reference picture with the clipped value. . A computer-readable recording medium for recording encoded picture data and related data of the encoded picture data,
Information for specifying an update scaling weight used for scaling of a residual value during update processing in decoding of encoded picture data from a plurality of update scaling weights is a predetermined data unit of the encoded picture data. And recording on the recording medium.

Images