JP4199712B2 - Decoded video quantization error reduction method and apparatus, decoded video quantization error reduction program used for realizing the quantization error reduction method, and computer-readable recording medium storing the program - Google Patents

Description

  The present invention is used in a video encoding system that performs encoding by determining a prediction mode of an encoding target block from among a plurality of prediction modes, or a video that decodes video encoded by the video encoding system Decoding video quantization error reduction method and apparatus used in decoding method, decoded video quantization error reduction program used for realizing quantization error reduction method, and computer-readable recording medium recording the program About.

  Currently, most of the mainstream video coding systems divide an image into a plurality of blocks and perform coding processing such as motion compensation prediction, DCT, and quantization in units of blocks.

  In general, in the moving picture coding system, when the coding bit rate is lowered, the quantization error increases and coding noise increases. In addition, due to the encoding process in units of blocks, the pixel change at the block boundary becomes discontinuous and block distortion occurs.

Therefore, many post-processing methods have been proposed in order to remove coding noise such as block distortion from the decoded image. Many of these methods are basically methods that focus on the continuity of block boundaries.
(I) Correction of DCT coefficient with block boundary continuity as a constraint condition (ii) Filtering to maintain block boundary continuity can be classified into two methods.

  First, a “DCT coefficient correction method with the block boundary continuity as a constraint (hereinafter referred to as a DCT coefficient correction method)” will be described.

  The occurrence of block distortion that does not exist in the original image is caused by the quantization error. For this reason, the continuity at the block boundary is used as a constraint, and the quantization error is reduced by correcting the DCT coefficient. The following non-patent document 1 shows an example.

  In the method described in Non-Patent Document 1, a low-pass filter is applied to the block boundary to determine the correction direction of the DCT coefficient. The correction width of the DCT coefficient is determined by the size of the quantization step size.

FIG. 6 shows the procedure. As shown in this process flow, the procedure is:
1. Filter block boundaries (step 100)
2. DCT is applied to the filtered signal (step 110).
3. It is examined how each DCT coefficient has changed before and after the filter (step 120).
4). The DCT coefficients before the filter are corrected according to the quantization step size with respect to the direction of change of each DCT coefficient (step 130).
5. Inverse DCT is applied to the corrected DCT coefficient (step 140).
It is composed by repeating.

  By repeating this process until the DCT coefficients converge (step 150), the quantization error can be reduced. In this method, since the processing target is a DCT coefficient, it acts on the pixels in the block.

  Next, the “block distortion removing filter for removing the edge of the block boundary” will be described.

  In this method, the DCT coefficient is not corrected to reduce the distortion, but the edge of the block boundary is removed by a low-pass filter. However, it is not always possible to remove block distortion by simply filtering the block boundaries. When a strong filter is applied, the block distortion is reduced, but the entire screen is blurred and the image quality is deteriorated. On the other hand, a weak filter cannot remove block distortion.

  Therefore, in ITU-T H.264 (hereinafter referred to as H.264, see Non-Patent Document 2), the intensity of block distortion is determined from the difference in the coding mode of adjacent blocks and the amount of change in pixels near the block boundary. Is used to switch the strength of the filter.

FIG. 7 shows the procedure. As shown in this process flow, the procedure is:
1. Collect coding mode information of neighboring blocks (step 200).
2. The block boundary strength is calculated from the difference in coding mode (step 210).
3. A change threshold is calculated from the quantization parameter (step 220).
4). The change amount of the pixel in the vicinity of the block boundary is compared with the threshold value, and if the change amount is equal to or greater than the threshold value, the process is terminated (step 230).
5. If the change amount is not greater than or equal to the threshold, a filter is selected from the block boundary strength (step 240).
6). The selected filter is applied to the decoded signal (step 250).
Consists of.

  Originally, the purpose of the block distortion removal filter is not to reduce the quantization error, but to eliminate visual interference caused by block distortion, but by adaptively performing filtering like the method adopted in H.264. , The SN ratio is improved.

In this manner, the quantization error can be reduced by correcting the DCT coefficient or applying an adaptive filter to the block boundary.
Narita Kiyoto, Toki, Kimoto Ihiko, Tanimoto Masayuki, “Repetition Reduction Method of Quantization Noise in DCT Coded Images,” IEICE A, Vol.J79-A, No.1, pp.69-76 , Jan. 1996. ITU-T Rec. H.264, "Advanced video coding for generic audiovisual services," 2003.

  In the prior art, quantization error can be reduced by using the continuity of the block boundary and correcting the DCT coefficient, or by adaptively switching the filter in consideration of differences in edge strength and coding mode. It is.

  However, each of these conventional techniques has problems. The problems of each method are described below.

  First, the problem of the DCT coefficient correction method will be described.

  In the DCT coefficient correction method, DCT coefficient correction is repeatedly performed. However, if this iteration is repeated until the DCT coefficient converges, there is a possibility of excessive processing, and there is a possibility that the image quality will deteriorate.

  Next, the problem of the block distortion removal filter will be described.

  In the block distortion removal filter, the filter strength is controlled in accordance with the coding conditions in which block distortion is likely to occur and the amount of change in pixels near the block boundary. However, since the estimation of the block distortion amount is not always accurate, it may be excessively blurred depending on the image. That is, depending on the input signal and encoding conditions, the image quality may be degraded by the block distortion removal filter. This tendency is particularly strong when the bit rate is high.

  The present invention has been made in view of such circumstances, and decodes a video by a video encoding method that employs a configuration in which a prediction mode of a block to be encoded is determined from a plurality of prediction modes and encoded. In addition, when decoding video with a video decoding scheme that employs a configuration of decoding video encoded with that video encoding scheme, the magnitude of the quantization error of the decoded video can be estimated, An object of the present invention is to provide a new technique that can reduce a quantization error of a decoded video by making it possible to correct a decoded signal based on the above.

  In the prior art, since the quantization error is reduced with the continuity of the block boundary as a constraint, there is a possibility that the video is corrected more than necessary.

  Therefore, in the present invention, the decoding side re-searches the prediction mode using the same method as the prediction mode search method used at the time of encoding, estimates the size of the quantization error from the re-search result, and performs decoding. Correct the image.

  In video encoding schemes such as H.264, encoding target blocks are predicted from encoded frames or encoded signals in the same frame to improve encoding efficiency. In particular, in H.264, when intra-frame encoding is performed, a prediction signal is generated from information of an already-encoded block of an encoding target block, and a prediction residual signal is encoded. Since multiple prediction modes are prepared, the prediction mode with the lowest cost is determined from the prediction residual signal power and the overhead code amount of the prediction mode (code amount of code for indicating the prediction mode) at the time of encoding. Then, the encoding process is performed.

  When the re-search result of the prediction mode by the decoded video is different from the prediction mode at the time of encoding, it is an influence due to the quantization error included in the decoded video. Therefore, when the prediction mode at the time of encoding and the prediction mode of the re-search result are different, the quantization error can be estimated.

Now, a video signal having a block size of N × N is considered as an N × N-dimensional vector. The original signal vector is <S>, the prediction signal vector is <Pa>, the prediction residual signal vector is <Da>, the encoded prediction residual signal vector is <D'a>, and the quantization error vector is <E>. Then, the decoded signal vector <S ′> is
<S '> = <D'a> + <Pa>
= (<Da> + <E>) + <Pa>
= (<Da> + <Pa>) + <E>
= <S> + <E>
It is expressed by the formula.

  As described above, the decoded signal vector <S ′> can be expressed by the sum of the original signal vector <S> and the quantization error vector <E>. That is, it can be said that the decoded signal vector <S ′> has moved by the quantization error vector <E>.

  FIG. 1 shows an example in a two-dimensional pixel space. The broken line in the figure is a boundary line where the prediction mode is switched, and the prediction mode at the time of encoding is a, and the prediction mode at the time of re-search is b.

  The reason that the prediction mode is changed by the re-search is that the original signal vector <S> is moved from the region a to the region b by the quantization error vector <E>. Therefore, the quantization error vector <E> is a difference vector (prediction signal difference) between the prediction signal vector <Pa> that is the representative point of the prediction mode a and the prediction signal vector <Pb> that is the representative point of the prediction mode b. Vector <Pab>).

  In the present invention, a component in the direction of the prediction signal difference vector <Pab> of the quantization error vector <E> is calculated to obtain a correction signal vector <C> of the decoded signal vector <S ′>.

FIG. 2 shows the procedure of the present invention. As shown in this processing flow, in the present invention,
1. Extraction of prediction mode information at the time of encoding (step 10)
2. Decoding processing is performed to generate a decoded signal vector <S ′> (step 11).
3. The prediction mode is searched again for the decoded signal vector <S ′> (step 12).
4). The change of the prediction mode information is confirmed, and if there is no change, the process is terminated (step 13).
5. If there is a change in the prediction mode information, a correction signal vector <C> is calculated (step 14).
6). The corrected signal vector <C> is added to the decoded signal vector <S ′> (step 15).
Execute the procedure.

  Here, the correction signal vector <C> is a product of the prediction signal difference vector <Pab> and the correction coefficient α.

  The correction coefficient α includes a method of determining a constant in advance and a method of calculating according to the distance from the decoded signal to the boundary line of both prediction modes. Further, by adjusting the correction coefficient α, it is possible to change the strength of the correction.

The decoded signal vector <S ″> after correction is
<S ″> = <S ′> + α × <Pab>
It is calculated according to the following formula.

As described above, the decoded video quantization error reduction apparatus of the present invention can be used in a video encoding system that performs encoding by determining a prediction mode of an encoding target block from a plurality of prediction modes. When used in a video decoding system that decodes video encoded by the video encoding system, and (a) when generating a decoded signal, the same method as that used for encoding is used. Re-search means for re-searching the prediction mode; (b) comparison means for comparing the prediction mode used for encoding with the prediction mode searched by the re-search means; and (c) the prediction mode is changed by the comparison means. If it is detected, by obtaining a difference of the prediction signal of the two prediction modes, and generating means for generating a compensation signal in accordance with the product of the difference signal obtained as the correction coefficient, generation of (d) generating means Corrected signal is decoded signal By adding, configured to an adding means for reducing the quantization error of the decoded video.

In the case employing this arrangement, generation means uses or generates a correction signal by using the correction coefficient which defines Me pre, a correction coefficient which varies with the distance from the decoded signal to the boundary of the two prediction modes A correction signal may be generated.

  Then, when using a correction coefficient that changes according to the distance from the decoded signal to the boundary between the two prediction modes, the generation unit uses the encoding cost considering the overhead code amount of the prediction mode and uses the correction cost according to the distance. A correction coefficient may be calculated.

  The decoding video quantization error reduction method of the present invention realized by the operation of each of the above processing means can also be realized by a computer program. This computer program is provided by being recorded on an appropriate recording medium. The present invention is realized by being provided via a network, installed when the present invention is implemented, and operating on a control means such as a CPU.

  According to the present invention, when decoding a video with a video encoding scheme that adopts a configuration in which a prediction mode of a block to be encoded is determined from among a plurality of prediction modes and performing encoding, When decoding video using a video decoding method that decodes the encoded video, the quantization error of the decoded video can be estimated, and the decoded signal can be corrected based on that. Therefore, the quantization error of the decoded video can be reduced.

  Next, the present invention will be described in detail according to embodiments.

  FIG. 3 shows an example of an embodiment of a video encoding apparatus 1 comprising the present invention.

  As shown in this figure, a video encoding apparatus 1 comprising the present invention includes an encoded data decoding unit 10, a prediction signal generation unit 11, a memory 12, a decoded video generation unit 13, and a prediction mode search unit 14. A correction signal generation unit 15 and a corrected decoded video generation unit 16. Here, a portion surrounded by a broken line shown in the figure is a portion for performing a conventional decoding process.

  The encoded data decoding unit 10 receives encoded data encoded by an encoding mechanism (not shown) as an input, and generates a prediction residual signal (prediction residual signal vector <D′ a shown in FIG. 1) from the encoded data. >) And prediction mode information (prediction mode information a shown in FIG. 1) are decoded.

  The prediction signal generation unit 11 uses the prediction mode information a provided from the encoded data decoding unit 10 and generates a prediction signal (prediction signal vector <Pa shown in FIG. 1) from the encoded decoded video stored in the memory 12. >).

  The decoded video generation unit 13 adds the prediction residual signal given from the encoded data decoding unit 10 and the prediction signal generated by the prediction signal generation unit 11 to add a decoded video (decoded signal vector < S ′>) is generated and stored in the memory 12.

  The prediction mode search unit 14 regenerates prediction mode information (prediction mode information b shown in FIG. 1) from the decoded video generated by the decoded video generation unit 13 and the encoded decoded video stored in the memory 12. Explore.

  The correction signal generation unit 15 encodes the prediction mode information a at the time of encoding given from the encoded data decoding unit 10, the prediction mode information b re-searched by the prediction mode search unit 14, and a prediction signal ( From the prediction signal vector <Pa> shown in FIG. 1 and the prediction signal (prediction signal vector <Pb> shown in FIG. 1) in the prediction mode information b, a correction signal (correction signal vector <P> shown in FIG. 1) is obtained. C>).

  The corrected decoded video generation unit 16 adds the decoded video generated by the decoded video generation unit 13 and the correction signal generated by the correction signal generation unit 15 to reduce the quantization error (shown in FIG. 1). The decoded signal vector <S ″>) is generated.

  In the video encoding device 1 having the present invention configured as described above, the prediction mode search unit 14 includes the decoded video generated by the decoded video generation unit 13 and the encoded decoded video stored in the memory 12. From this, the prediction mode information b is searched again.

  In response to the re-search result, the correction signal generation unit 15 matches the prediction mode information a at the time of encoding given from the encoded data decoding unit 10 with the prediction mode information b re-searched by the prediction mode search unit 14. When determining whether or not they match, it is determined that a quantization error exists, and a correction signal is generated based on the difference between the prediction signals in the two prediction mode information a and b. To do.

  In response to the generation of the correction signal, the corrected decoded video generation unit 16 adds the decoded video generated by the decoded video generation unit 13 and the correction signal generated by the correction signal generation unit 15 to reduce the quantization error. The decoded video is generated and output.

  In this manner, the video encoding apparatus 1 including the present invention can reduce the quantization error by correcting the decoded video with the prediction signal difference vector.

  In the above description, the prediction mode is used, which means information necessary for generating a prediction signal. Therefore, in the case of inter-frame prediction encoding with motion compensation, the prediction mode information indicates all information necessary for generating a prediction signal, such as a motion vector and a reference frame number. And the presence or absence of a prediction mode change is compared about all these information.

  In addition to post-processing on the decoding side, the present invention can perform processing immediately before storing the decoded signal in the memory on both the encoding side and the decoding side, similar to the H.264 block removal filter. It is.

  Next, the present invention will be described in detail according to specific examples.

  In this embodiment, an intra-frame coding scheme that can use a plurality of prediction modes is assumed. The prediction block size is fixed, and the prediction mode information is only the prediction mode number.

Prediction mode selection, taking into account the overhead code quantity g _i (a overhead costs of the prediction mode i) of each prediction mode i,
j _i = | <D _i > | ² + g _i
= | (<S> − <P _i >) | ² + g _i
<D _i >: Prediction residual signal vector
<S>: Original signal vector
<P _i >: This is performed by calculating the coding cost j _i of each prediction mode according to the equation of the prediction signal vector and selecting the prediction mode with the lowest coding cost.

Here, the vector length | D _i | is calculated by the L2 norm shown in the following equation.

FIG. 4 shows the procedure of this embodiment. As shown in this processing flow, in this embodiment,
1. A prediction mode number at the time of encoding is extracted (step 20).
2. Decoding processing is performed to generate a decoded signal vector <S ′> (step 21).
3. The intra-frame prediction mode is searched again for the decoded signal vector <S ′> (step 22).
4). The change of the prediction mode number is confirmed, and if there is no change, the process is terminated (step 23).
5. When the prediction mode number is changed, a correction signal vector <C> is calculated (step 24).
6). The corrected signal vector <C> is added to the decoded signal vector <S ′> (step 25).
Execute the procedure.

  At this time, the correction signal vector <C> is obtained as a product of the prediction signal difference vector <Pab> and the correction coefficient α.

As a result, the corrected decoded signal vector <S ″> is
<S ″> = <S ′> + α × <Pab>
Thus, in this embodiment, the corrected decoded signal vector <S ″> is output.

  For the correction coefficient α, a method of determining a constant in advance may be used, but as described below, a method of calculating according to the distance from the decoded signal to the boundary line of both prediction modes may be used. .

If the overhead cost of the prediction mode number a at the time of encoding is g _a and the overhead cost of the prediction mode number b at the time of re-search is g _b , the decoded signal vector <S ′> is represented as shown in FIG. alpha ₀ × the addition of <Pab>, it is assumed that moving on to the border of the two prediction modes.

At this time, since the coding cost on the boundary line is equal in both prediction modes,
| <D ″ a> | ² + g _a = | <D ″ b> | ² + g _b
It becomes.

Here, as can be seen from FIG. 5, <D ″ a>, <D ″ b> and the decoded signal vector <S ″> (= <S ′) when moved on the boundary line between both prediction modes. > + Α ₀ × <Pab>)
<D''a> = <S''>-<Pa>
<D ″ b> = <S ″> − <Pb>
From the fact that
−2 <D′ a> · <Pab> − (2α ₀ +1) | <Pab> | ² −g _b + g _a = 0
However, “·” is an inner product of vectors.

From now on, α ₀ is
_{α 0 = {g a -g b} - | <Pab> | 2 -2 <D'a> · <Pab>} / 2 | <Pab> | 2
It becomes.

In general, since the original signal is often located closer to the prediction signal than on the boundary between prediction modes, β indicating a value larger than 1 is used as the correction coefficient, and α = β × α _0. Is preferably set. Here, when β = 2, the correction amount is twice the distance from the decoded signal vector <S ′> to the boundary line of both prediction modes.

  In this embodiment, the prediction block size is fixed, but it may be variable. As the block size changes, the dimensionality of the pixel space only changes. When correction is performed with different block sizes, correction may be performed in a larger block size, that is, in a pixel space with a higher number of dimensions. In addition, although intra-frame prediction coding is assumed, even in the inter-frame coding method, processing can be similarly performed by replacing prediction mode information with information necessary for prediction signal generation such as motion compensation mode and motion vector. .

  The present invention can be used for a video encoding apparatus that performs encoding by determining a prediction mode of an encoding target block from among a plurality of prediction modes. The present invention can be used in a video decoding device that decodes video encoded by the video encoding device.

It is a figure which shows the relationship between a decoded signal vector, an original signal vector, and a quantization error vector. It is a processing flow which this invention performs. 1 is an embodiment of a video encoding device provided with the present invention. It is the processing flow which a present Example performs. It is a figure explaining derivation | leading-out of a correction coefficient. It is a processing flow of a prior art. It is a processing flow of a prior art.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Video coding apparatus 10 Encoded data decoding part 11 Prediction signal generation part 12 Memory 13 Decoded video generation part 14 Prediction mode search part 15 Correction signal generation part 16 Correction decoding video generation part

Claims (10)

Used in video coding schemes that perform coding by determining the prediction mode of the encoding target block from multiple prediction modes, or used in video decoding schemes that decode video coded by the video coding scheme A method for reducing quantization error of decoded video,
Re-searching the prediction mode in the same manner as used for encoding when generating the decoded signal;
Comparing the prediction mode used for encoding with the prediction mode of the re-search results;
If the prediction mode is changed, we calculate the difference between the prediction signal of the two prediction modes, and generating a compensation signal in accordance with the product of the difference signal obtained correction coefficient,
Adding the correction signal to the decoded signal,
A feature of a method for reducing a quantization error of a decoded video. The method for reducing a quantization error of decoded video according to claim 1 ,
In the generation process, generating a correction signal using a predetermined correction coefficient,
A feature of a method for reducing a quantization error of a decoded video. The method for reducing a quantization error of decoded video according to claim 1 ,
In the generation process, generating a correction signal using a correction coefficient that changes according to the distance from the decoded signal to the boundary between both prediction modes,
A feature of a method for reducing a quantization error of a decoded video. The quantization error reduction method for decoded video according to claim 3 ,
In the process of generating, calculating a correction coefficient according to the distance using an encoding cost considering the overhead code amount of the prediction mode,
A feature of a method for reducing a quantization error of a decoded video. Used in video coding schemes that perform coding by determining the prediction mode of the encoding target block from multiple prediction modes, or used in video decoding schemes that decode video coded by the video coding scheme A decoding error reduction apparatus for decoded video,
Means for re-searching the prediction mode in the same way as the method used for encoding when generating the decoded signal;
Means for comparing the prediction mode used for encoding with the prediction mode of the re-search result;
If the prediction mode is changed, it calculates the difference between the prediction signal of the two prediction modes, means for generating a compensation signal in accordance with the product of the difference signal obtained correction coefficient,
Means for adding the correction signal to the decoded signal;
A feature of an apparatus for reducing quantization error of decoded video. The quantization error reduction apparatus for decoded video according to claim 5 ,
The generating means generates a correction signal using a predetermined correction coefficient,
A feature of an apparatus for reducing quantization error of decoded video. The quantization error reduction apparatus for decoded video according to claim 5 ,
The generating means generates the correction signal using a correction coefficient that changes in accordance with the distance from the decoded signal to the boundary between both prediction modes.
A feature of an apparatus for reducing quantization error of decoded video. The decoded video quantization error reduction device according to claim 7 ,
The generating means calculates a correction coefficient corresponding to the distance using an encoding cost considering the overhead code amount of the prediction mode.
A feature of an apparatus for reducing quantization error of decoded video. A decoded video quantization error reduction program for causing a computer to execute processing used to realize the decoded video quantization error reduction method according to any one of claims 1 to 4 . A computer-readable recording in which a decoded video quantization error reduction program for causing a computer to execute a process used to realize the decoded video quantization error reduction method according to any one of claims 1 to 4 is recorded. Medium.

Images