KR20110067539A - Method and apparatus for video coding and decoding using intra prediction - Google Patents

Abstract

PURPOSE: An intra predictive coding/decoding method and apparatus thereof are provided to perform parametric according to a prediction direction of an input image and to reduce computational complexity. CONSTITUTION: One or more pixel of a coding target block is determined as a prediction target pixel(S110). Compensating pixels are determined as prediction support pixels(S120). The compensating pixels are determined as prediction support pixels.

Description

Intra prediction encoding / decoding method and apparatus {METHOD AND APPARATUS FOR VIDEO CODING AND DECODING USING INTRA PREDICTION}

The present invention relates to an intra prediction encoding / decoding method and apparatus, and more particularly, to a technique for performing intra prediction on an input image using a prediction direction.

The present invention is derived from the research conducted as part of the IT source technology development project of the Korea Communications Commission and the Korea Institute of Information and Telecommunication Research and Development. [Task Management No .: 2008-F-011, Title: Development of Next-Generation DTV Core Technology (Standardization) -Next generation DTV core technology development.

In general, a video encoding / decoding apparatus having a hybrid prediction encoding structure performs encoding by dividing an input image into units of 16 × 16 macroblocks. In particular, the video compression standards ISO / IEC 14496-10 (MPEG-4 Advanced Video Coding) or H.264 and ISO / IEC 14496-10 Amendment 3 (MPEG-4 Scalable Video Coding) standards, etc. In-picture prediction is performed on the subblocks 4x4 and 8x8. In the case of performing intra prediction, since the residual signal is encoded rather than the original signal in the block, the intra coding performance efficiency is improved.

However, in the case of accommodating parametric prediction on a pixel-by-pixel basis, in the conventional encoding apparatus, when the number of pixel values used for prediction increases, computational complexity increases.

Accordingly, there is a need for a method capable of accommodating local changes in the predicted pixel value for which prediction is to be performed while reducing computational complexity.

The intra prediction encoding method may further include determining at least one pixel included in an encoding target block as a prediction target pixel based on a prediction direction based on a prediction direction, and a reconstructed pixel included in the input image. Determining reconstructed pixels, which are adjacent to an encoding target block and around a prediction direction, as prediction based pixels, determining training data sets based on the determined positions of the at least one prediction pixel and the prediction based pixels, Calculating a prediction parameter used to reduce a prediction error according to the prediction direction based on the determined training data sets, and a prediction direction for at least one prediction target pixel determined using the calculated prediction parameter and the determined prediction based pixels. To perform parametric predictions It can hamhal.

In this case, the determining of the prediction based pixels may include determining final prediction based pixels by interpolating a pixel value corresponding to an intermediate position between the prediction based pixels.

The determining of the prediction based pixels may include calculating an upper limit reference value and a lower limit reference value based on an average of the prediction based pixels, a standard deviation, and a quantization parameter of an encoding target block, and pixel values of the determined prediction based pixels. The method may include determining final prediction based pixels with pixels included between the calculated upper limit reference value and the lower limit reference value.

The determining of the prediction based pixels may include determining a similar pixel from the determined prediction based pixels, selecting one of the determined similar pixels as a representative pixel, and selecting a similar pixel except the representative pixel among the determined similar pixels. The method may include adding weights to the pixels, and determining final prediction based pixels using weighted similar pixels among the prediction based pixels, and weights.

In addition, the intra prediction decoding method may further include receiving an intra prediction encoded input image and targeting at least one block constituting the received input image, based on a prediction direction, at least one pixel included in the decoding target block. Determining as a prediction target pixel, determining decoding pixels, which are adjacent to a decoding target block among the decoded pixels included in the input image and located around the prediction direction, as prediction based pixels, the at least one determined prediction pixel Determining training data sets based on the position of prediction based pixels, calculating a prediction parameter used to reduce a prediction error along a prediction direction based on the determined training data sets, and calculating the calculated prediction parameters and the determined At least one prediction table determined using prediction based pixels Can include a prediction direction for the pixel performs the decoding in parametric screen.

According to the present invention, not only the computational complexity can be reduced by performing parametric prediction on the input image according to the prediction direction, but also the intra prediction can be performed in detail.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart provided to explain an intra prediction encoding method.

Referring to FIG. 1, the encoding apparatus may determine a prediction target pixel from an encoding target block according to a prediction direction (S110). Here, the encoding target block may be a block to be encoded among a plurality of blocks included in the input image. In one example, the input image may include expansion macro blocks larger than 16 × 16.

In this case, the encoding target block may be one of each block or sub-blocks below it, and the prediction target pixel may be at least one of a plurality of pixels included in the encoding target block. For example, the prediction target pixel may be expressed as in Equation 1 below.

Here, K is the number of prediction target pixels in each prediction direction. In this case, the prediction target pixel may be one or more.

In more detail, as shown in FIG. 2, the encoding apparatus may determine, as the prediction target pixel 230, pixels located on the prediction direction line among pixels included in the encoding target block. Here, the prediction direction may be preset to one or more of diagonal left downward, diagonal right downward, diagonal left upward, diagonal right upward, horizontal left direction, horizontal right direction, vertical upward direction, vertical downward direction, and zero direction.

In addition, the encoding apparatus may determine an optimal direction for prediction of an encoding target block by using an encoding method in an advanced video coding (AVC) screen. Then, the determined optimal direction can be used as the prediction direction.

Subsequently, the encoding apparatus may determine reconstructed pixels adjacent to the encoding target block and located around the prediction direction as prediction based pixels (S120). Here, the pixels reconstructed after encoding is completed among the plurality of pixels included in the input image may be used as the prediction based pixels. For example, the prediction based pixel may be expressed as Equation 2 below.

Here, N is the number of prediction based pixels, and n and N may be differently determined according to the prediction direction or k in Equation 1.

More specifically, as shown in FIG. 2, the encoding apparatus may determine, as prediction base pixels 210, reconstructed pixels located close to a prediction direction line among reconstructed pixels located around the prediction base pixel 230. In general, as N increases, the complexity of the encoder and the decoder increases. As described above, by adaptively selecting the prediction-based pixels according to the prediction direction, the computation amount and the prediction performance may be improved during encoding and decoding.

In this case, the encoding apparatus may determine final prediction based pixels by using interpolation between prediction based pixels, an average between prediction based pixels, and similar pixels (S130).

For example, when interpolation between prediction based pixels is used, and as shown in FIG. 3, when the size of an encoding target block is 4 × 4 and the prediction direction is diagonally left downward, the encoding apparatus is a middle part of each of the prediction based pixels 210. The final prediction based pixels 330 may be determined by interpolating pixel values of the pixels 330 corresponding to positions.

As such, when interpolation between prediction based pixels is used, the number of prediction based pixels N may be reduced from 10 to 6 as not all prediction based pixels are used. Accordingly, the computational complexity due to the prediction may be further reduced, and the prediction performance may also be improved because the noise pixels are flattened by interpolation.

As another example, when the average between prediction-based pixels is used, and the size of the encoding target block is 4 × 4 and the prediction direction is diagonally left downward, as shown in FIG. Can be calculated

Then, the encoding apparatus may calculate the upper limit reference value and the lower limit reference value using the calculated average value, standard deviation, and quantization parameter of the encoding target block. Here, the quantization parameter of the encoding object block may be obtained through quantization of the encoding object block.

For example, the encoding apparatus may calculate the upper limit reference value T _high and the lower limit reference value T _low according to Equation 3 below.

T _low <y ⁽⁰⁾ (n) <T _high , n = 1,… , N

^{_{T low = mean {y (0}} ) (n)} - scale (QP) · deviation {y (0) (n)}

T _high = mean {y ⁽⁰⁾ (n)} + scale (QP) · deviation {y ⁽⁰⁾ (n)}

Here, mean {y ⁽⁰⁾ (n)} is an average value of prediction based pixels, scale (QP) is a standard deviation of prediction based pixels, and deviation {y ⁽⁰⁾ (n)} is a quantization parameter of a block to be encoded. .

In more detail, the encoding apparatus may calculate a lower limit reference value by subtracting the product of the standard deviation of the prediction-based pixels and the quantization parameter from the average of the prediction-based pixels. In addition, the encoding apparatus may calculate a lower limit reference value by adding a product of the standard deviation of the prediction-based pixels and the quantization parameter to the average of the prediction-based pixels.

Then, the encoding apparatus may determine, as the final prediction based pixel, pixels that exceed a lower limit reference value among the prediction base pixels and fall below the upper limit reference value.

In addition, pixels having similar pixel values among the prediction-based pixels may be weighted instead of being included as prediction-based pixels to give the same effect to reduce the number of prediction-based pixels. That is, as shown in FIG. 3, when the size of the encoding target block is 4 × 4, the prediction direction is diagonally left downward, and similar pixels are used, the encoding apparatus has a difference in pixel value among the prediction-based pixels that is less than the predetermined reference pixel value. A set of similar pixels can be determined.

Then, the encoding apparatus may select any one of the determined similar pixels as the representative pixel. In addition, the encoding apparatus may assign a weight to only the representative pixel to use as a prediction based pixel. Then, since the remaining similar pixels are not used as prediction based pixels, the number of prediction based pixels may be reduced, thereby reducing complexity.

In this case, the encoding apparatus may determine the final prediction based pixel by using reconstructed pixels except the similar pixel, the representative pixel, and the weight among the prediction based pixels.

In one example, the encoding apparatus may determine a similar pixel from the prediction based pixels based on Equation 4 below.

y ⁽⁰⁾ (i) -y ⁽⁰⁾ (j) | <Th, i, j = 1,... , N… (One)

Weight (i) = the number of pixels equation (1)

Here, Th may be preset as a reference pixel value, and weight (i) is a weight.

According to Equation 4, the encoding apparatus includes a pixel in which an absolute value (| y ⁽⁰⁾ (i) -y ⁽⁰⁾ (j) |) of the difference between prediction-based pixels is less than the reference pixel value Th. Can be determined to be similar pixels. Then, the encoding apparatus may select any one of the crystal-like pixels as the representative pixel. In this case, the encoding apparatus may calculate an average value of the similar pixels and select the average value as the representative pixel.

In addition, the encoding apparatus may select a similar pixel positioned in the middle of the similar pixels as the representative pixel. Then, the encoding apparatus may add a weight as the number of the similar pixels except the selected representative pixel among the similar pixels.

For example, when the number of prediction-based pixels is 10 and five similar pixels are determined, the encoding apparatus may add a weight of 4 to four similar pixels except the representative pixel. Then, the encoding apparatus may determine the final prediction based pixels using six pixels and weights excluding the similar pixels to which the weight is added in the prediction based pixel. In this case, the added weight may be used to calculate the prediction parameter.

As described above, when the prediction base pixel is determined to include many reconstructed pixels adjacent to the prediction direction line, prediction performance may be improved. In particular, the encoding apparatus may determine reconstructed pixels in as many directions as possible among up / down / left / right of the encoding target block as prediction based pixels.

 In this case, when the encoding target block is located at the slice boundary or adjacent to the non-reconstructed block, the encoding apparatus predicts pixels among the reconstructed pixels included in the input image that are close to the direction vector in the prediction direction of the encoding target block. The base pixels may be selected.

For example, as illustrated in FIG. 4, there may be pixels U that have not been reconstructed among the pixels U and A adjacent to the prediction direction line. Then, the encoding apparatus may determine the pixels S adjacent to the direction vector among the reconstructed pixels and the reconstructed pixels A adjacent on the prediction direction line as the final prediction based pixels. In this case, as shown in FIG. 5, when the size of an encoding target block is 4 × 4, the encoding apparatus may differently determine prediction based pixels according to various prediction directions.

Subsequently, the encoding apparatus may determine training data sets based on the geometric positional relationship between the determined prediction target pixel and the final prediction based pixel (S140).

For example, as shown in FIG. 9, the encoding apparatus may virtually position a prediction target pixel with respect to each pixel position of a specific region of pixels that are already reconstructed. After the prediction target is virtually located, the encoding apparatus may determine the training data set using the prediction based pixels and the prediction target pixel template.

As another example, the encoding apparatus may determine the training data sets to include reconstructed pixels close to a distance from the prediction-based pixels determined in operation S120. In this case, the training data sets may be determined by Equation 5 below. In addition, the template representing the determined training data sets is shown in FIG. 6.

Here, M may represent the total number of training data sets for a given prediction direction.

In more detail, the encoding apparatus may determine the training data sets to have a pixel value similar to the pixel value of the prediction-based pixels and the pixel value of the prediction target pixels. Here, the determined training data sets may be used when calculating a prediction parameter.

In addition, the encoding apparatus may determine the training data sets such that the template-matched pixel value error between the prediction-based pixels and the corresponding points is reduced. In this case, the training data sets may be determined differently according to the prediction direction.

That is, when the prediction direction is Z, Z training data sets may be determined. As a result, since it is not necessary to determine a training data set for each of the prediction target pixels, computational complexity may be reduced. Here, Z can be any constant.

In addition, the encoding apparatus may generate a common template as shown in FIG. 7. That is, when the prediction direction is Z, the encoding apparatus may generate one common template such that prediction-based pixels corresponding to each of the Z prediction directions are included in a predetermined number or more. Then, the encoding apparatus may determine training data sets having the same geometry as the generated common template.

In operation S150, the encoding apparatus may calculate a prediction parameter for the pixels included in the determined training data sets so that the prediction error according to the prediction direction is reduced. In this case, the encoding apparatus may calculate a prediction parameter using an affine prediction technique or a linear prediction technique.

For example, when the linear prediction technique is used, the prediction target pixel may be predicted by Equation 6 below.

 According to Equation 6, the k-th prediction target pixel may be predicted using the m-th training data set.

In this case, Equation 6 may be expressed in a matrix-vector format as shown in Equation 7 below.

Here, is an additive noise generated when the k-th prediction target pixel is predicted from the m-th training data.

More specifically, the encoding apparatus includes a least square error (LSE) technique, a weighted least square error (WLSE) technique, a best linear unbiased estimator (BLUE), and Prediction parameters can be calculated using the Linear Minimum Mean Square Error (LMMSE) technique. In this case, the prediction parameter may include an optimization cost function and a solution.

For example, when the least square error method is used, the encoding apparatus may calculate an optimization cost function and a solution, respectively, as shown in Equation 9 below.

In addition, when using the weight-least-error method, the encoding apparatus may calculate an optimization cost function and a solution, respectively, as shown in Equation 9 below.

In addition, in the case of using an optimal linear bias estimator technique, the encoding apparatus may calculate an optimization cost function and a solution, respectively, as shown in Equation 10 below.

In addition, when the linear minimum mean square error method is used, the encoding apparatus may calculate an optimization cost function and a solution, respectively, as shown in Equation 11 below.

는 부가 잡음

에 대한 공분산 행렬(Covariance Matrix)이 고,

는 파라미터 a_k의 사전(Priori) 확률을 이용하여 산출된 파라미터 a_k의 평균 벡터이고,

는 파라미터 a_k의 공분산 행렬이다.here,

Is an additive noise

Is the covariance matrix for,

And is the mean vector of a parameter a _k calculated by using the dictionary of the parameter a _k (Priori) probability,

Is the covariance matrix of the parameter a _k .

,

,

를 산출하는 경우, 부호화 장치는 트레이닝 데이터 집합과 예측 기반 화소들 간의 유사도를 이용하여 가중치 행렬 w_k을 산출할 수 있다. 일례로, 가중치 행렬 w_k은 아래의 수학식 12에 의해 산출될 수 있다.More specifically, the least squares error technique

,

,

When calculating, the encoding apparatus may calculate the weight matrix w _k using the similarity between the training data set and the prediction based pixels. For example, the weight matrix w _k may be calculated by Equation 12 below.

은 아래의 수학식 13에 의해 산출될 수 있다.Also, the study matrix for additive noise

May be calculated by Equation 13 below.

및 공분산 행렬

는 아래의 수학식 14에 의해 산출될 수 있다.Also, the average vector of parameter a _k

And covariance matrix

May be calculated by Equation 14 below.

Here, as samples used to calculate the prediction parameters, parameters for blocks that are close to the current encoding target block among blocks that are already encoded may be used. In addition, parameters for the training data set may be used as a sample used to calculate the prediction parameters.

및 공분산 행렬

는 위의 수학식 14에 의해 산출될 수 있다.In this case, when P parameter samples are set, the average vector of the parameters a _k

And covariance matrix

May be calculated by Equation 14 above.

Subsequently, the encoding apparatus may perform parametric prediction on the prediction target pixel by using the calculated prediction parameter and the prediction based pixels (S160).

In this case, the pixel value of the prediction target pixel may be determined according to a mathematical function relationship between prediction based pixels, as shown in Equation 15. For example, the encoding apparatus may determine the pixel value of the prediction target pixel by performing parametric prediction based on Equation 15 below.

Then, the pixel value of the prediction target pixel determined by Equation 15 may be expressed by Equation 16 below.

In addition, the encoding apparatus may perform encoding only on an error between the determined pixel value of the prediction target pixel and the actual pixel value. Through this, encoding efficiency may be improved.

8 is a flowchart provided to explain an intra prediction prediction method.

First, the decoding apparatus may receive an intra prediction encoded image. Here, the intra prediction coded image may include a plurality of blocks. Also, the plurality of blocks may each include a plurality of sub blocks.

Then, the decoding apparatus may determine at least one pixel included in the decoding target block as the prediction target pixel based on the prediction direction (S810). Here, the decoding target block may be any one of the plurality of blocks on which decoding is not performed.

 Subsequently, the decoding apparatus may determine decoded pixels adjacent to the decoding target block and located around the prediction direction as prediction based pixels (S820). Here, the decoded pixels may be pixels that are already decoded by decoding among a plurality of pixels included in the intra prediction coded image. In this case, since the feature of determining the prediction based pixels is the same as that described in operation S120 of FIG. 1, a detailed description thereof will be omitted.

In operation S840, the decoding apparatus may calculate a prediction parameter used to predict the decoding target block based on the determined training data sets. In this case, the decoding apparatus may calculate a prediction parameter such that the prediction error according to the prediction direction is reduced. Here, since the feature for calculating the prediction parameter is the same as that described in step S150, detailed description thereof will be omitted.

 Subsequently, the decoding apparatus may perform parametric prediction on the prediction target pixel by using the calculated prediction parameter and the prediction based pixels (S850). In this case, the decoding apparatus may perform parametric prediction on the prediction target pixel in the prediction direction. In this way, intra prediction decoding may be performed on the prediction target pixel.

Up to now, the encoding and decoding method for input image consisting of a plurality of macroblocks and performing intra prediction on the plurality of macroblocks has been described. However, this is an exemplary embodiment. May include subblocks of. Accordingly, parametric prediction encoding and decoding may be performed on pixels included in the plurality of subblocks according to a prediction direction.

In addition, the encoding apparatus may perform transform encoding on the input image. That is, the encoding apparatus may selectively apply one or more transform kernels according to the block size to the residual signal obtained through intra prediction. Then, the decoding apparatus may perform transform decoding on the intra prediction encoded image.

Also, the encoding apparatus may compare the intra prediction encoding results of the subblocks during the intra prediction. The encoding apparatus may determine an optimal encoding mode for the subblock. Then, the decoding apparatus may perform decoding in an optimal encoding mode corresponding to each subblock. In this case, the decoded image is an image received by intra prediction encoding, and may be composed of a plurality of blocks, and the plurality of blocks may include a plurality of sub blocks.

In addition, in the above description for the convenience of description, the equation is described as an example, but this is an embodiment, the present encoding and decoding apparatus should not be limited to the equation (1 to 16).

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above embodiments, and those skilled in the art to which the present invention pertains various modifications and variations from such descriptions. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined not only by the claims below but also by the equivalents of the claims.

1 is a flowchart provided to explain an intra prediction encoding method.

2 to 5 are diagrams provided to explain a method of determining prediction based pixels.

6 and 9 are diagrams provided to explain a method of determining training data sets.

7 is a view provided to explain a common template.

8 is a flowchart provided to explain an intra prediction prediction method.

Claims (10)

Determining at least one pixel included in an encoding target block as a prediction target pixel based on a prediction direction, based on an input image including a plurality of blocks; Determining reconstructed pixels adjacent to the encoding target block among reconstructed pixels included in the input image and located around a prediction direction as prediction based pixels; Determining training data sets based on the determined at least one prediction target pixel and the position of the prediction based pixels; Calculating a prediction parameter used to reduce a prediction error according to the prediction direction based on the determined training data sets; And Performing parametric prediction in the prediction direction on the determined at least one prediction target pixel by using the calculated prediction parameter and the determined prediction based pixels. Intra prediction encoding method comprising a. The method of claim 1, The determining of the prediction based pixels may include: Determining final prediction based pixels by interpolating a pixel value corresponding to an intermediate position between the prediction based pixels. Intra prediction encoding method comprising a. The method of claim 1, The determining of the prediction based pixels may include: Calculating an upper limit reference value and a lower limit reference value based on the average of the prediction-based pixels, a standard deviation, and a quantization parameter of the encoding target block; And Determining the final prediction based pixels from pixels among the determined prediction based pixels whose pixel values are included between the calculated upper limit reference value and the lower limit reference value. Intra prediction encoding method comprising a. The method of claim 1, The determining of the prediction based pixels may include: Determining a similar pixel from the determined prediction based pixels; Selecting one of the determined similar pixels as a representative pixel; Adding weights to similar pixels except the representative pixel among the determined similar pixels; And Determining the final prediction based pixels using the weighted similar pixels among the prediction based pixels, and the weights. Intra prediction encoding method comprising a. The method of claim 1, Determining the training data sets, And determining the prediction based pixels and the reconstructed pixels close thereto by the training data sets. The method of claim 1, Determining the training data sets, Generating a template having the same geometry as the prediction based pixels and the at least one prediction target pixel Including, And determining the training data set such that a template-matched pixel value error between the prediction based pixels and the generated template is reduced. The method of claim 1, Computing the prediction parameter, And calculating the prediction parameter from the determined training data sets using a linear prediction technique or an affine prediction technique. The method of claim 1, Computing the prediction parameter, And calculating the prediction parameter based on a difference between prediction based pixels corresponding to pixels included in the training data sets among the prediction based pixels. The method of claim 1, Each of the plurality of blocks includes a plurality of sub blocks, Performing the prediction, The intra prediction encoding method according to claim 1, wherein the intra prediction is performed on the plurality of subblocks. Receiving an intra prediction coded input image; Determining at least one pixel included in a decoding object block as a prediction target pixel based on a prediction direction, for a plurality of blocks constituting the received input image; Determining decoded pixels, which are adjacent to the decoding target block among the decoded pixels included in the input image and located around a prediction direction, as prediction based pixels; Determining training data sets based on the determined at least one prediction target pixel and the position of the prediction based pixels; Calculating a prediction parameter used to reduce a prediction error according to the prediction direction based on the determined training data sets; And Performing parametric intra-picture decoding on the determined at least one prediction target pixel in the prediction direction by using the calculated prediction parameter and the determined prediction based pixels. In-screen decoding method comprising a.

Images