Apparatus for encoding an image

ES 2 575 381 T3

DESCRIPTION

Intra-prediction decoding device

The present invention relates to a decoding apparatus, and more specifically, to an apparatus that decodes an intra-prediction mode, adaptively decodes a prediction block and a residual block according to the intra-prediction mode, and thus generates a reconstructed block.

In image compression procedures such as Moving Picture Expert Group (MPEG) -1, MPEG-2, MPEG-4, and H.264 / MPEG-4 Advanced Video Coding (AVC), an image is divided into macroblocks to encode an image. The respective macroblocks are then encoded using an inter-prediction or an intra-prediction.

In intra-prediction, a block of a current image is encoded not using a reference image, but using pixel values ​​spatially adjacent to the current block. A low distortion intra-prediction mode is selected by comparing with an original macroblock using the adjacent pixel values. Next, using the selected intra-prediction mode and adjacent pixel values, the prediction values ​​of the current block are calculated. And a difference between the prediction values ​​and the pixel values ​​of the original current block is calculated and then encoded through a transform encoding, quantization, and entropy encoding. The intra-prediction mode is also coded.

Intra-prediction modes are generally classified into a 4 x 4 intra-prediction mode, an 8 x 8 intra-prediction mode, and a 16 x 16 intra-prediction mode for the luminance components and chrominance components.

In the 16 x 16 intra-prediction mode according to the related art there are four modes, a vertical mode, a horizontal mode, a direct current (DC) mode and a flat mode.

In the 4 x 4 intra-prediction mode according to the related art there are nine modes, a vertical mode, a horizontal mode, a CC mode, a bottom-left diagonal mode, a bottom-right diagonal mode, a right vertical, one left vertical mode, one horizontal-up mode, and one horizontal-down mode.

Each prediction mode is indexed according to the frequency of use of the respective modes. The portrait mode, which is mode 0, shows the highest chance that it will be used most frequently to perform intra-prediction on a target block, and the horizontal-up mode which is mode 8 shows the highest chance that used the most infrequently.

According to H.264 standards, a current block is encoded using a total of 13 modes, that is, 4 modes of the 4 x 4 intra-prediction mode and 9 modes of the 16 x 16 intra-prediction mode. A bit stream of the current block is generated according to an optimal mode between these modes.

However, when some or all of the values ​​of the pixels adjacent to the current block do not exist or are not already encoded, it is impossible to apply some or all of the intra-prediction modes. Also, when a difference between adjacent reference pixels is large, a difference between a prediction block and an original block becomes large. Therefore, a new technique is required to reduce the difference between the original block and the generated prediction block based on the positions of the reference pixels used to generate the prediction block.

European patent application EP2557797 by Lee Jin Ho et al. presented 04/11/2011 discloses an apparatus for decoding, the apparatus comprising: an entropy decoder configured to restore quantized residual coefficients and intra-prediction information; a prediction mode decoder configured to restore an intra-prediction mode based on the intra-prediction information; a residual signal decoder configured to restore a residual signal using the intra-prediction mode; a reference pixel generator configured to generate the reference pixels corresponding to the unavailable reference pixels and to adaptively filter the reference pixels according to the intra-prediction mode; a prediction block generator configured to generate the prediction pixels using the reference pixels determined by the intra-prediction mode; a prediction block filter configured to adaptively filter some of the prediction pixels using the intra-prediction mode; and an image reconstructor configured to generate a reconstructed image using the prediction pixels and the residual signal.

The present invention relates to a decoding apparatus for effectively reconstructing an encoded image with a high compression efficiency by generating or reconstructing a prediction block close to an original image.

One aspect of the present invention provides a decoding apparatus, including: an entropy decoder configured to restore quantized residual coefficients and intra-prediction information; a prediction mode decoder configured to restore an intra-prediction mode based on the

ES 2 575 381 T3 intra-prediction information; a residual signal decoder configured to restore a residual signal using the intra-prediction mode; a reference pixel generator configured to generate the reference pixels corresponding to the unavailable reference pixels and adaptively filter the reference pixels using the intra-prediction mode; a prediction block generator configured to generate the prediction pixels using the reference pixels determined by the intra-prediction mode; a prediction block filter configured to adaptively filter some of the prediction pixels using the intra-prediction mode; and an image reconstructor configured to generate a reconstructed image using the prediction pixels and the residual signal.

A decoding apparatus according to the present invention generates reference pixels and adaptively filters the reference pixels in order to generate a prediction block similar to an original block. In addition, by generating or modifying the prediction block that uses the reference pixels that are not used to generate a prediction block, the prediction block can be reconstructed similar to an original block, and an image compression can be improved.

FIG. 1 is a block diagram of a moving picture coding apparatus in accordance with the present invention.

Figure 2 is a block diagram of an intra predictor according to the present invention.

Figure 3 is a conceptual diagram showing directional intra-prediction modes in accordance with the present invention.

Fig. 4 is a flow chart illustrating an intra-prediction mode encoding procedure of a current prediction unit performed in a prediction mode encoder in accordance with the present invention.

Figure 5 is a block diagram of a decoding apparatus in accordance with the present invention.

Hereinafter, the various embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to the exemplary embodiments disclosed below, as it can be implemented in various ways. Therefore, many other modifications and variations of the present invention are possible, and it should be understood that within the scope of the disclosed concept, the present invention may be practiced other than as specifically described.

For image coding, each image consists of one or more segments, and each segment is made up of a plurality of coding units. Since an image of a high definition (HD) grade or higher has many uniform regions, a compression of the image can be improved by encoding the image with encoding units of various sizes.

The coding units according to the present invention have a quadruple tree structure and can be divided hierarchically using depth information. A larger size coding unit is referred to as a larger coding unit (LCU), and a smaller size coding unit is referred to as a smaller coding unit (SCU). Information regarding the LCU and SCU can be included in a Sequence Parameter Set (SPS) and transmitted.

An LCU consists of one or more encoding units. The LCU is in the form of a recursive coding tree in order to include a division structure of the coding units. When the LCU is not divided into four coding units, the coding tree may consist of information indicating that the LCU is not divided and a coding unit. When the LCU is divided into four coding units, the coding tree may consist of information indicating that the LCU is divided and four sub-coding trees. Similarly, each sub-coding tree has the same structure as the LCU coding tree. However, an SCU-sized coding unit is not divided into coding units.

Meanwhile, each coding unit in the coding tree undergoes an intra-prediction or inter-prediction in units of the coding unit itself or a sub-partition. A unit in which an intra-prediction or an inter-prediction is made is referred to as a prediction unit. A prediction unit size can be 2N x 2N or N x N in an intra-prediction. A prediction unit size can be 2N x 2N, 2N x N, N x 2N, or N x N in an interprediction. In this document, 2N indicates the horizontal and vertical lengths of the encoding unit.

Meanwhile, a prediction unit for intra-prediction may not be a square. For example, a square coding unit can be divided into four hN x 2N or four 2N x hN for intra-prediction. In this case, the distance between a reference pixel and a pixel of a prediction block for intra-prediction is reduced, such that the prediction efficiency can be improved. This intra-prediction procedure is a short-range intra-prediction (SDIP).

ES 2 575 381 T3

A coding unit includes the prediction mode information and the size information (part_mode) of the prediction units within the coding unit. To improve coding efficiency, prediction mode information and size information can be combined and coded together. In this case, each encoding unit includes a joint encoded prediction type (web_type).

A coding unit includes the additional information necessary to generate a prediction block from each prediction unit and a residual signal. Additional information is defined per prediction unit in the encoding unit. In intra-prediction, the additional information includes the coded intra-prediction information. In inter prediction, the additional information includes the encoded motion information. The motion information includes a motion vector and a reference image index.

A residual signal is included in each coding unit. A residual signal includes a transform tree, a luminance residual signal carrier, and two chrominance residual signal carriers. Residual signal carriers include residual information encoded in one or more transform units. The largest size of the transform unit is equal to or less than the size of the encoding unit. The transform unit can be the same size as the largest transform unit or a sub-transform unit of the largest transform unit.

The transform tree includes the information indicating a transform unit division structure for the residual signal included in the coding unit. In addition, the transform tree includes the information that indicates whether or not a residual signal of each transform unit is 0.

The residual signal carrier carries the residual information encoded in the transform units corresponding to the information indicating the division structure in the transform tree in units of encoding units.

Although the above description has only been made of a prediction unit obtained by dividing a coding unit equally, uneven division is also possible. In other words, to understand a residual signal, it may be more efficient to unevenly divide an image signal in a specific direction according to an image boundary and perform an intra or inter prediction.

The simplest adaptation mode is to divide a coding unit into two blocks using a straight line in order to extract the statistical dependence of a prediction region from the local topography. A boundary of an image corresponds to the straight line and is divided. In this case, the divisible addresses may be limited to a predetermined number. For example, a procedure of dividing a block may be limited to four directions of horizontal, vertical, diagonal up and diagonal down directions. Also, the division may be limited to only the horizontal and vertical directions. The number of divisible addresses can be three, five, seven, and so on. The number of divisible addresses can vary according to a block size. For example, for an encoding unit of a large size, the number of divisible addresses can be increased relatively.

In an inter prediction, when a coding unit is divided into two prediction units for one more adaptive prediction, motion estimation and motion compensation should be performed in each of the prediction units. After the prediction blocks are derived from the respective two divided prediction units of the coding unit, the two prediction blocks can be added to generate the prediction block having the same size as the coding unit. In this case, to reduce the difference between the pixel values ​​on both sides of a division boundary of the encoding unit size prediction block, pixels located at the division boundary can be filtered. The prediction block can be generated in such a way that the prediction blocks corresponding to the respective prediction units overlap, and the overlap boundary portion can be smoothed to generate the prediction block.

FIG. 1 is a block diagram of a moving picture coding apparatus in accordance with the present invention.

Referring to FIG. 1, a moving image coding apparatus 100 in accordance with the present invention includes an image splitter 110, a transformer 120, a quantizer 130, a scanner 131, an entropy encoder 140, an intra predictor 150, an inter predictor 160, an inverse quantizer 135, an inverse transformer 125, a post processor 170, an image storage 180, a subtractor 190, and an adder 195.

Image splitter 110 analyzes an input video signal to divide each LCU of an image into one or more units each of which has a predetermined size, determines the prediction mode of each encoding unit, and determines the size of the prediction unit for each coding unit. Image splitter 110 sends the prediction unit to be coded to intra predictor 150 or inter predictor 160 according to the prediction mode. That is, the image splitter 110 sends the prediction units to be encoded to the subtractor 190.

ES 2 575 381 T3

Transformer 120 transforms a residual block that is a residual signal between an original block of an input prediction unit and a prediction block generated by intra-predictor 150 or inter-predictor 160. Preferably, the residual block consists of a unit of coding. The residual block can be divided into optimal transform units and transformed. A type of transform matrix can be adaptively determined according to the prediction mode (intra or inter) and the intra-prediction mode. The transform unit can be transformed by horizontal and vertical one-dimensional (1D) transform matrices. In inter prediction, a predetermined transform matrix is ​​applied. In intra-prediction, there is a high possibility that the residual block has vertical directionality when the intra-prediction mode is horizontal. Therefore, an integer matrix based on the discrete cosine transform (DCT) is applied to the vertical direction, and a discrete sine transform (DST) or an integer matrix based on the Karhunen Loeve transform (KLT) is applied ) to the horizontal direction. When the intra-prediction mode is vertical, a DST or KLT-based integer matrix is ​​applied to the vertical direction, and a DCT-based integer matrix is ​​applied to the horizontal direction. Alternatively, in intra-prediction, the transform matrix can be adaptively determined according to the size of the transform unit.

Quantizer 130 determines a quantization step size for each coding unit in order to quantize the transform coefficients of the residual block transformed by the transform matrix. The quantization step size is determined by the encoding unit size equal to or greater than a predetermined size. The default size can be 8x8 or 16x16. Using the determined quantization step size and a determined quantization matrix according to a prediction mode, the transform coefficients are quantized. Quantizer 130 uses the quantization step sizes of the adjacent coding units of the current coding unit as a quantization step size predictor of the current coding unit. Quantizer 130 sequentially searches for a left coding unit, an upper coding unit, and an upper left coding unit of the current coding unit, determines the quantization stage size predictor of the current coding unit using the sizes quantization step of one or more available coding units, and transmits a difference to the entropy encoder 140.

When a segment is divided into coding units, there may be none of a left coding unit, an upper coding unit, and an upper left coding unit of the current coding unit. On the other hand, there may be a previous coding unit of an LCU in a coding order. Therefore, coding units adjacent to the current coding unit and the previous coding unit may be candidates. In this case, the priority can be administered sequentially given for 1) the left coding unit of the current coding unit, 2) the upper coding unit of the current coding unit, 3) the upper left coding unit of the current coding unit and 4) the previous coding unit of the current coding unit. The sequence may vary, or the upper left coding unit may be omitted.

The quantized transform block is provided to inverse quantizer 135 and scanner 131.

The scanner 131 scans the quantized transform block coefficients and converts the quantized transform block coefficients to 1D quantized coefficients. A coefficient scanning procedure is determined according to the prediction mode and the intra-prediction mode. Also, the coefficient scanning procedure can be determined differently according to the size of the transform units. The scanner 131 determines whether or not the quantized coefficient block is divided into a plurality of subsets based on a current transform unit size. When the size of the transform unit is larger than a first reference size, the quantized transform block is divided into a plurality of subsets. The first reference size can be 4x4 or 8x8.

Scanner 131 determines a scan pattern to be applied to the quantized transform block. In inter-prediction, only a predetermined scan pattern can be applied (for example, a zigzag scan). In intra-prediction, a determined scan pattern can be applied according to the intra-prediction mode. The scan pattern may vary according to a directional intra-prediction mode. Zigzag scanning applies to non-directional intra-prediction modes. A non-directional mode can be a direct current (DC) mode or a flat mode. The quantized coefficients are scanned analyzed in an inverse direction.

When the quantized coefficients are divided into the plurality of subsets, the same scan pattern is applied to the quantized coefficients in each subset. The plurality of subsets consists of a main subset and one or more residual subsets. The main subset is located on an upper left side and includes a coefficient of CC and the one or more residual subsets that cover a region other than the main subset.

Zigzag scanning can be applied to scan subsets. The subsets can be scanned starting with the main subset and then with the residual subsets in a direct direction, or they can be scanned in a reverse direction. A scan pattern to scan the subsets can be set the same as a scan pattern to scan the quantized coefficients in the subsets.

ES 2 575 381 T3 subsets. In this case, the scan pattern for the subsets is determined according to the intra-prediction mode. Meanwhile, an encoder transmits information capable of indicating a position of the last non-zero quantized coefficient of the transform unit to a decoder. The encoder also transmits information capable of indicating a position of the last non-zero quantized coefficient in each subset to the decoder.

The inverse quantizer 135 inverse quantizes the quantized transform coefficients. The inverse transformer 125 reconstructs a residual block of the spatial domain of the inversely quantized transform coefficients. The adder 195 generates a reconstruction block by adding the residual block reconstructed by the inverse transformer 125 and the prediction block of the intra predictor 150 or the inter predictor 160.

Post processor 170 performs an unblocking filtering procedure to remove the blocking disturbance generated in a reconstructed image, an adaptive compensation application procedure to supplement a difference between the reconstructed image and the original image per pixel, and a procedure of adaptive loop filter to supplement a difference between the reconstructed image and the original image in one encoding unit.

The unblocking filtering procedure can be applied to a boundary between the prediction units and between the transform units. The default size can be 8 x 8. The unblocking filtering procedure includes a step of determining the limit to be filtered, a step of determining the limit filtering force to be applied to the limit, a step of determining whether or not to apply an unblocking filter, and a step of selecting a filter to be applied to the limit when it is determined to apply the unblocking filter.

Applying the unblocking filter or not is determined according to i) if the limit filtering force is or not greater than 0 and ii) if a value that indicates the difference between the limit pixels of the two blocks (the block P and the block Q) adjacent to the limit to be filtered is or is not less than a first reference value determined according to a quantization parameter.

There can be two or more filters. When an absolute value of a difference between two pixels adjacent to the block boundary is equal to or greater than a second reference value, a weak filter is selected. The second reference value is determined by the quantization parameter and the limit filtering force.

The adaptive compensation application method is intended to reduce a difference (distortion) between a pixel in an image subjected to the unblocking filter and the original pixel. An image or segment can be divided into a plurality of compensation regions, and one compensation mode per compensation region can be determined. There are four edge compensation modes, two band compensation modes, and a no-offset mode. According to each compensation mode, the pixels in each compensation region are classified into a predetermined number of classes, and the compensation corresponding to the classified class is added to the pixel. In the case of an edge compensation mode, a class of a current pixel is determined by comparing the current pixel value with the pixel values ​​of two or more pixels adjacent to the current pixel.

The adaptive loop filter procedure can be performed based on a value obtained by comparing an original image with a reconstructed image to which the deblocking filtering procedure or the adaptive compensation applying procedure is applied. An adaptive loop filter (ALF) is detected through a Laplacian activity value based on a 4 x 4 block. The determined ALF can be applied to all the pixels included in a 4 x 4 block or in an 8 x 8 block. Whether or not to apply an ALF can be determined according to each encoding unit. A size and coefficients of a loop filter can vary according to each coding unit. Information indicating whether the ALF applies to each encoding unit, filter coefficient information, filter shape information, and so on, can be included in the segment header and transmitted to the decoder. In the case of a chrominance signal, whether or not to apply the ALF can be determined in the imaging units. Unlike luminance, the loop filter can be rectangular in shape.

Image storage 180 receives post-processing image data from post-processor 170, and stores the reconstructed image in image units. An image can be an image in a frame or in a field. Image storage 180 has a buffer (not shown) capable of storing a plurality of images.

Interpredictor 160 performs motion estimation using one or more reference images stored in image storage 180, and determines a reference image index indicating the reference images and a motion vector. According to the reference image index and the motion vector, a prediction block corresponding to a prediction unit to be encoded is extracted from a reference image selected from a plurality of reference images stored in the storage 180 of images and the extracted prediction block is output.

Intra-predictor 150 performs intra-prediction coding using a reconstructed pixel value in an image that includes a current prediction unit. The intra predictor 150 receives the current prediction unit to be encoded,

ES 2 575 381 T3 selects one of a predetermined number of intra-prediction modes, and performs the intra-prediction. The predetermined number of intra-prediction modes depends on a size of the current prediction unit. The intra-predictor 150 adaptively filters the reference pixels used to generate the intra-prediction block. When some of the reference pixels are not available, it is possible to generate the reference pixels at the unavailable positions using the available reference pixels.

Entropy encoder 140 encodes transform coefficients quantized by quantizer 130, intra-prediction information received from intra-predictor 150, motion information received from inter-predictor 160, and so on.

Figure 2 is a block diagram of an intra predictor 150 in accordance with the present invention.

Referring to FIG. 2, the intra predictor 150 includes a prediction unit receiver 141, a reference pixel generator 142, a prediction block generator 143, a prediction block post processor 144, and a determiner 145 of prediction modes and a prediction mode encoder 146.

Prediction unit receiver 141 receives prediction unit input from image splitter 110. Prediction unit receiver 141 transmits size information in the received prediction unit to prediction mode determiner 145 and generator 142 of reference pixels, and transmits the prediction unit to the reference pixel generator 142 and to the prediction block generator 143.

The reference pixel generator 142 determines whether the reference pixels of the currently received prediction unit are available. The reference pixels of the current prediction unit used for intra-prediction consist of a corner reference pixel placed at (x = -1, y = -1), 2L upper reference pixels placed at (x = 0, ..., 2L-1, y = -1), and 2M left reference pixels placed at (x = 0, y = 0, ..., and 2M-1). In this case, L is the width of the current prediction unit, and M is the height of the current prediction unit.

When the reference pixels are not available or insufficient, the reference pixels are generated.

When reference pixels are not available, reference pixels are generated with a default value.

When some of the reference pixels are not available, it is determined whether the unavailable reference pixels exist in a single direction of the available pixels or between the available pixels.

When the unavailable reference pixels exist in only one direction from the available pixels, a reference block is generated by copying the value of the available pixel closest to the unavailable pixel. For example, when the current prediction unit is placed on an upper boundary of an image or segment, the corner reference pixel and the upper reference pixels are not available. Therefore, in this case, the corner reference pixel and the upper reference pixels can be generated by copying a reference pixel placed at (x = -1, y = 0) which is the closest position. Alternatively, the reference pixels can be generated using one available reference pixel from the closest position and one or more available reference pixels. For example, when the corner reference pixel placed at (x = - 1, y = 0) and the reference pixels at positions (x = 0, ..., and L-1, y = -1) are available and reference pixels at positions (x = L, ..., 2L- 1, y = -1) are not available, reference pixels at unavailable positions can be generated using the change in difference between a reference pixel value at a position (x = L-1, y = -1) and a corner reference pixel value or other reference pixel value.

When the unavailable reference pixels exist among the available pixels, the reference pixels are generated using two available pixels p and q adjacent to the unavailable reference pixels. For example, when the corner reference pixel and L top reference pixels placed at (x = 0, ..., and L-1, y = -1) are not available, the reference pixels that exist between the reference pixel p at a position (x = -1, y = 0) and the reference pixel q at a position (x = L, y = -1) using the reference pixels p and q.

The generated reference pixel values ​​can be obtained by rounding an average of the reference pixel p and the reference pixel q. Furthermore, the reference pixel values ​​can be generated using the change in the difference between the pixel values ​​of the reference pixel p and the reference pixel q. In this case, the reference pixel values ​​can be generated by a linear interpolation determined according to the positions corresponding to the pixels to be generated or by using a weighted average of the two reference pixels.

Meanwhile, when a plurality of prediction units are on an upper side of the current prediction unit, there is a high possibility that a difference between the boundary pixels present on both sides of a boundary between two of the upper prediction units is greater than a difference between adjacent pixels in each higher prediction unit. This results from an error caused by a quantization parameter. The error is most likely to occur in directional intra-prediction modes where a prediction block is generated using the two adjacent reference pixels.

ES 2 575 381 T3

In particular, the modes (mode numbers 3, 6 and 9) that have a direction of 45 ° with respect to the horizontal or vertical direction of Figure 3 are most seriously affected. In portrait and landscape modes (mode numbers 0 and 1), one pixel is used to generate a prediction pixel from the prediction block, and thus portrait and landscape modes are slightly affected.

For this reason, a filter (the uniformity filter) is applied to the reference pixels in the 3, 6, and 9 directional intraprediction modes, and it is not applied to the reference pixels in the vertical and horizontal intra-prediction modes. . In the CC mode between the non-directional intra-prediction modes, the filter is also not applied to the reference pixels. For these modes, it can be determined to apply the filter or not regardless of the size of the current prediction unit.

In the directional intra-prediction modes that exist between intra-prediction mode 3, 6, or 9 and the horizontal or vertical intra-prediction mode, the filter (the uniformity filter) can be applied adaptively to the pixels of reference according to the size of the prediction unit. It is preferable to increase a probability of applying the filter (the uniformity filter) when the direction of the directional intra-prediction mode is relatively closer to the direction of the intra-prediction mode which has the direction of 45 °. Specifically, when a first directional mode is closer in directionality to the intra-prediction mode having the 45 ° direction than a second directional mode, if a filter is applied to the reference pixels in the second directional mode, the filter it also applies to reference pixels in the first directional mode. On the other hand, if the filter is applied to the reference pixels in the first directional mode, the filter may or may not be applied to the reference pixels in the second directional mode.

There is a high possibility that a change in the difference between the pixels in a prediction unit of a large size is less than a change in the difference between the pixels in a prediction unit of a small size. Therefore, the number of directional modes in which the filter is applied may increase or the stronger filter may be applied as the size of the prediction unit increases. On the other hand, when the prediction unit becomes smaller than a specific size, the filter may not be applied.

For an example, in intra-prediction mode 3, 6, or 9, a first filter can be applied to the reference pixels of a prediction unit having a size equal to or less than a first size, and a second filter can be applied. that is stronger than the first filter to the reference pixels of a prediction unit that is larger than the first size. The first size can vary according to the directional prediction modes.

In another example, in intra-prediction mode 5 that exists between vertical intra-prediction mode and intra-prediction mode 6, a filter cannot be applied to a prediction unit having a size equal to or less than a second size, the first filter can be applied to the reference pixels of a prediction unit having a size greater than the second size and equal to or less than a third size, and the second filter can be applied to a prediction unit having a size larger than the third size. The second size and the third size can vary according to the directional prediction modes.

The first filter can be a 3-lead filter [1, 2, 1] or a 5-lead filter [1, 2, 4, 2, 1]. The second filter has a greater uniformity effect than the first filter.

The prediction block generator 143 generates a prediction block using the reference pixels determined by the intra-prediction mode.

In the directional intra-prediction mode, the corresponding reference pixels vary according to the intra-prediction mode. For example, in portrait mode, L top reference pixels placed at (x = 0, ..., and L-1, y = -1) are used, and in landscape mode, L left reference pixels are used placed at (x = -1, y = 0, ..., and L-1).

In non-directional intra-prediction mode, the corner pixel, the top L reference pixels placed at (x = 0, ..., and L-1, y = -1) and the left L reference pixels are used placed at (x = -1, y = 0, ..., and L-1). The non-directional intra-prediction mode is CC mode or flat mode.

In flat mode, a reference pixel from a prediction block is generated using the corner reference pixel, a left reference pixel, and an upper reference pixel. When a reference pixel to be generated is placed at (a, b), a prediction pixel X (a, b) is generated using corner reference pixel C (x = -1, y = -1), one pixel upper reference pixel T (x = a, y = -1) and a left reference pixel L (x = -1, y = b). Specifically, X (a, b) can be L (x = -1, y = b) + T (x = a, y = -1) - C (x = -1, y = -1).

In the intra-prediction mode that exists on the right side of the vertical mode (the mode number is 0) of Figure 3, the possibility that the differences between the pixels of a lower left region of the generated prediction block and the Corresponding pixels of the original prediction unit will increase if the prediction block is generated using only the top reference pixels. However, when a prediction block is generated using the upper reference pixels and the left reference pixels, the difference can be reduced. This effect is greater in intra-prediction mode 6. Also in the intra-prediction modes that

ES 2 575 381 T3 exist in the horizontal mode (the mode number is 1) of Fig. 3, the same procedure can be applied, and the effect is greater than in the intra-prediction mode 9.

Therefore, in intra-prediction mode 6 or 9, a prediction pixel can be generated using the corresponding upper interpolation reference pixel (for example, positioned at 45 ° from the position angle) and a reference pixel of left interpolation. The prediction pixel can be generated by linear interpolation of a top interpolation reference pixel and the left interpolation reference pixel or by using an average that is rounded. Also, in a predetermined number of intra-prediction modes adjacent to mode 6 or 9, a prediction block can be generated using the left reference pixels and the upper reference pixels. For example, in intra-prediction mode 6 or in a mode belonging to the predetermined number of (eg four) prediction modes adjacent to mode 6, the left reference pixels and the upper reference pixels can be used to generate a prediction block. In this case, to reduce complexity, the above-mentioned procedure cannot be applied in intra-prediction modes having mode numbers greater than a predetermined mode number (eg 9 or 17). Also, the procedure can be applied to only a current prediction unit having a size equal to or greater than a predetermined size. The default size is 8x8 or 16x16.

The prediction block post processor 144 adaptively filters the prediction block generated by the prediction block generator 143. To reduce the differences between a reference pixel and the pixels adjacent to the reference pixel, the prediction block post processor 144 adaptively filters some or all of the pixels adjacent to the reference pixel block according to the mode of intra -prediction. Pixels adjacent to the reference pixel exist in the prediction block.

In flat mode, the pixels adjacent to the reference pixel are generated using the reference pixel, and therefore no filter is applied.

In CC mode, an average of reference pixels is used, and therefore a filter is applied. Different types of filter can be used according to the size of the prediction unit. For a prediction unit of a large size, a filter that is the same as that used in a prediction unit of a small size or a strong filter that has a large uniformity effect can be used.

Prediction mode determiner 145 determines the intra-prediction mode of the current prediction unit using the reference pixels. The prediction mode determiner 145 may determine an intra-prediction mode with the anticipated minimum encoded bits of a residual block as an intra-prediction mode of the current prediction unit. In this case, the residual block is generated using a prediction block corresponding to each intra-prediction mode or a post-processing prediction block.

The prediction mode encoder 146 encodes the intra-prediction mode of the current prediction unit using the intra-prediction modes of the prediction units adjacent to the current prediction unit.

FIG. 4 is a flow chart illustrating an intra-prediction mode encoding procedure of a current prediction unit performed in the prediction mode encoder 146 in accordance with the present invention.

First, the candidates of the intra-prediction mode of a current prediction unit are searched (S110). A top intra-prediction mode and a left intra-prediction mode of the current prediction unit may be the candidates for the intra-prediction mode. A corner intra-prediction mode can be added as well, or any other mode can be added according to the upper intra-prediction mode and the left intra-prediction mode.

When there are a plurality of upper prediction units of the current prediction unit, the plurality of higher prediction units are scanned in a predetermined direction (for example, from right to left) to determine the intra-prediction mode of a first unit. Prediction available as a left intra-prediction mode. Alternatively, among a plurality of available prediction units, the intra-prediction mode of an available prediction unit having the lowest number of intra-prediction mode can be set as a higher intra-prediction mode.

The corner intra-prediction mode can be set as a prediction mode of a prediction unit adjacent to the upper right side or the upper left side of the current prediction unit. Alternatively, the corner intra-prediction mode can be set as a first available intra-prediction mode obtained by scanning the intra-prediction modes adjacent to the upper left side, upper right side, and lower right side of the prediction unit. current in a predetermined order. The default order is the top left side, the bottom right side, and the top right side. Alternatively, two (upper right side and upper left side) or three (upper right side, upper left side and lower left side) corner intra-prediction modes can be added as the candidates of the current prediction unit intra-prediction mode. .

ES 2 575 381 T3

Next, it is determined whether or not the intra-prediction mode of the available intra-prediction mode candidates is changed (S120).

When it is determined to change the intra-prediction mode, the available intra-prediction mode candidate is changed (S130).

Specifically, when the mode number of an available intra-prediction mode candidate is equal to or greater than the number of allowable intra-prediction modes for the current prediction unit, the available intra-prediction mode candidate becomes in one of the permissible modes. The number of allowable modes may vary according to a current prediction unit size. For example, when the current prediction unit size is 4 x 4, the intra-prediction mode of the available intra-prediction mode candidate becomes one of the nine modes (mode 0 to mode 8) or all 18 modes.

Next, an intra-prediction candidate list of the current prediction unit is constructed (S140). Candidates can be listed in a numerical order of modes. Alternatively, candidates can be listed in order of frequency, and intra-prediction candidates having the same frequency can be listed in numerical order of modes. When the intra-prediction mode candidates have the same mode number, the intra-prediction candidates are removed from the list except one.

Next, it is determined whether the intra-prediction mode of the current prediction unit is the same as any one of the intra-prediction mode candidates in the constructed list (S150).

When the intra-prediction mode of the current prediction unit is the same as one of the candidates of the intra-prediction mode, the information indicating that the intra-prediction mode is the same as one of the candidates of the intra-prediction mode and a candidate index as an intra-prediction information (S160).

When the intra-prediction mode of the current prediction unit is not the same as any of the intra-prediction mode candidates, a mode change value is calculated to change the intra-prediction mode of the current prediction unit ( S170). The mode change value is the number of the intra-prediction mode candidates having an intra-prediction mode number that is not greater than the intra-prediction mode number of the current prediction unit. The mode change value is obtained by comparing the intra-prediction mode values ​​in the list.

Next, the intra-prediction mode of the current prediction unit is changed using the mode change value (S180). The changed intra-prediction mode is determined as the intra-prediction mode of the current prediction unit. The changed intra-prediction mode of the current prediction unit is transmitted to the entropy encoder 140.

FIG. 5 is a block diagram of an intra-prediction decoding apparatus 200 in accordance with the present invention.

The intra-prediction decoding apparatus 200 according to the present invention includes an entropy decoder 210, a residual signal decoder 220, a prediction mode decoder 230, a reference pixel generator 240, a block generator 250 prediction block filter 260 and image reconstructor 270.

The entropy decoder 210 extracts the quantized residual coefficients from a received bit stream, and transmits the quantized residual coefficients and a transform unit size to the residual signal decoder 220 in units of transform units. Furthermore, the entropy decoder 210 transmits the intra-prediction information and a size of a prediction unit to be decoded to the prediction mode decoder 230.

The residual signal decoder 220 converts the quantized residual coefficients into an inverse quantized block of a two-dimensional (2D) matrix. For this conversion, one of a plurality of scan patterns is selected. The transform block scan pattern is determined based on at least one of the prediction mode and the intra-prediction mode. A reverse scan operation is the same as a reverse operation procedure of the scanner 131 of FIG. 1. For example, when a size of the current transform unit to be decoded is larger than a first reference size, the quantized residual coefficients are scanned inversely to form a plurality of subsets according to the scan pattern, and a inverse quantized block having the size of the transform unit using the plurality of subsets. On the other hand, when the size of the current transform unit to be decoded is not larger than the first reference size, the inverse quantized residual coefficients are scanned to generate an inverse quantized block having the same size of the transform unit. according to the scan pattern.

The prediction mode decoder 230 reconstructs the intra-prediction mode from the current prediction mode based on the intra-prediction information and the size information in the current prediction unit.

ES 2 575 381 T3 received from entropy decoder 210. The received intra-prediction information is restored through a reverse procedure of the procedure shown in Figure 4.

The reference pixel generator 240 generates reference pixels not available from the current prediction unit, and adaptively filters the reference pixels according to the current prediction unit intra-prediction mode received from the decoder 230. of prediction modes. A reference pixel generation procedure and a reference pixel filtering procedure are the same as that of the reference pixel generator 142 in the intra predictor 140 of FIG. 2.

Specifically, it is determined whether the reference pixels of the current prediction unit are available. The reference pixels of the current prediction unit used for intra-prediction consist of a corner reference pixel placed at (x = -1, y = -1), 2L upper reference pixels placed at (x = 0, ..., y 2L1, y = -1), and 2M left reference pixels placed at (x = 0, y = 0, ..., and 2M-1). In this case, L is a width of the current prediction unit, and M is a height of the current prediction unit.

When the reference pixels for generating a prediction block are not available or insufficient, reference pixels are generated.

When all the reference pixels are not available, the reference pixels are generated with a default value.

When some of the reference pixels are not available, it is determined whether there are any reference pixels that are not available in a single direction of the available pixels or between the available pixels.

When there are unavailable reference pixels in only one available pixel direction, the reference pixels are generated by copying the value of an available pixel closest to the unavailable pixel. For example, when the current prediction unit is placed on an upper boundary of an image or segment, the corner reference pixel and the upper reference pixels are not available. Therefore, in this case, the corner reference pixel and the upper reference pixels can be generated by copying a reference pixel placed at (x = -1, y = 0). Alternatively, the reference pixels can be generated using one available reference pixel from the closest position and one or more available reference pixels. For example, when the corner reference pixel that has a position (x = -1, y = -1) and the reference pixels at the positions (x = 0, ..., and L-1, y = - 1) are available and the reference pixels at the positions (x = L, ..., 2L-1, y = -1) are not available, the reference pixels at the unavailable positions can be generated using the change in the difference between a reference pixel at a position (x = L-1, y = -1) and a corner reference pixel value or other reference pixel value.

When the missing reference pixels exist among the available pixels, the reference pixels are generated using two available pixels p and q adjacent to the unavailable reference pixels. For example, when the corner reference pixel and the top L reference pixels placed at (x = 0, ..., and L-1, y = 1), are unavailable reference pixels that exist between the pixel of reference p at a position (x = -1, y = 0) and reference pixel q at a position (x = L, y = -1) can be generated using reference pixels p and q.

The generated reference pixel values ​​can be obtained by rounding an average of the reference pixel p and the reference pixel q. Furthermore, the reference pixel values ​​can be generated using the change in the difference between the pixel values ​​of the reference pixel p and the reference pixel q. In this case, the reference pixel values ​​can be generated by linear interpolation according to the positions corresponding to the generated pixel values ​​or by using a weighted average of the two reference pixels.

Meanwhile, when a plurality of prediction units are on an upper side of the current prediction unit, there is a high possibility that a difference between the boundary pixels present on both sides of a boundary between two of the upper prediction units is higher than a difference between adjacent pixels in each higher prediction unit. This results from an error caused by a quantization parameter. The error is most likely to occur in directional intra-prediction modes where a prediction block is generated using two adjacent reference pixels.

In particular, the modes (mode numbers 3, 6 and 9) having a direction of 45 ° with reference to a horizontal or vertical direction of Figure 3 are the most severely affected. In the vertical and horizontal intra-prediction modes (mode numbers 0 and 1), one pixel is used to generate a prediction block, and therefore the vertical and horizontal intra-prediction modes are slightly affected.

For this reason, a filter (the uniformity filter) is applied to the reference pixels in the 3, 6, and 9 directional intraprediction modes, and it is not applied to the reference pixels in the vertical and horizontal intra-prediction modes. . In DC mode outside of non-directional modes, the filter is also not applied. For these modes, it can be determined whether or not to apply the filter regardless of the size of the current prediction unit.

ES 2 575 381 T3

In the directional intra-prediction modes that exist between intra-prediction mode 3, 6, or 9 and the horizontal or vertical intra-prediction mode, the filter (the uniformity filter) can be applied adaptively to the pixels of reference. It is preferable to increase a probability of applying the filter (the uniformity filter) as the direction of the directional intra-prediction mode that is relatively closer to the direction of the intra-prediction mode that has the direction of 45 °. Specifically, when a first directional intra-prediction mode is closer in directionality to the intra-prediction mode having the 45 ° direction than a second directional intra-prediction mode, if a filter is applied to the second intra-prediction mode, directional prediction, the filter is also applied to the first directional intra-prediction mode. On the other hand, if the filter is applied to the first directional intra-prediction mode, the filter may or may not apply to the second directional intra-prediction mode.

There is a high possibility that a change in the difference between the pixels in a prediction unit of a large size is less than a change in the difference between the pixels in a prediction unit of a small size. Therefore, the number of directional modes in which the filter is applied can be increased or the stronger filter can be applied when the size of the prediction unit increases. On the other hand, when a prediction unit gets smaller than a specific size, the filter may not be applied.

For an example, in intra-prediction mode 3, 6 or 9, which has the direction of 45 °, a first filter can be applied to a prediction unit having a size equal to or less than a first size, and can be applied a second filter that is stronger than the first filter to a prediction unit that is larger than the first size. The first size can vary according to directional prediction modes.

For another example, in the intra-prediction mode 5 that exists between the vertical intra-prediction mode and the intra-prediction mode 6, which has the 45 ° direction, a filter cannot be applied to a prediction unit that has a size equal to or less than a second size, the first filter can be applied to a prediction unit that has a size greater than the second size and equal to or less than a third size, and the second filter can be applied to a prediction unit having a size larger than the third size. The second size and the third size can vary according to the directional prediction modes.

The first filter can be a 3-lead filter [1, 2, 1] or a 5-lead filter [1, 2, 4, 2, 1]. The second filter has a greater uniformity effect than the first filter.

The prediction block generator 250 generates a prediction block according to the intraprediction mode of the current prediction unit received from the prediction mode decoder 230. A procedure for generating the prediction block is the same as that of the prediction block generator 142 in the intra predictor 140 of FIG. 2.

That is, in the directional intra-prediction mode, the corresponding reference pixels vary according to the intra-prediction modes. For example, in portrait mode, the top L reference pixels placed at (x = 0, ..., and L-1, y = -1) are used, and in landscape mode, the L pixels of left references placed at (x = -1, y = 0, ..., and L-1).

In non-directional intra-prediction modes, the corner pixel, the top L reference pixels placed at (x = 0, ..., and L-1, y = -1) and the L reference pixels are used lefts placed at (x = -1, y = 0, ..., and L-1). The non-directional intra-prediction modes are CC mode and flat mode.

In flat mode, a reference pixel from a prediction block is generated using the corner reference pixel, a left reference pixel, and an upper reference pixel. When a reference pixel to be generated is placed in (a, b), the reference pixel of the prediction block X (a, b) is generated using the corner reference pixel C (x = -1, y = -1 ), an upper reference pixel T (x = a, y = -1) and a left reference pixel L (x = -1, y = b). Specifically, X (a, b) can be L (x = -1, y = b) + T (x = a, y = -1) - C (x = -1, y = -1).

In the intra-prediction mode that exists on the right side of the vertical mode (mode number 0) of Figure 3, there is the possibility that a difference between the pixels of a lower left region of the generated prediction block and the pixels of the original prediction unit if the prediction block is generated using only the top reference pixels. However, when a prediction block is generated using the upper reference pixels and the left reference pixels for various modes between the modes, the difference can be reduced. This effect is greater in intra-prediction mode 6. Furthermore, in the intra-prediction modes existing in the horizontal mode (mode number 1) of Fig. 3, the same procedure can be applied, and the effect is greater in the intra-prediction mode 9.

Therefore, in prediction mode 6 or 9, a corresponding upper interpolation reference pixel (eg, positioned 45 ° from the prediction pixel) and a left interpolation reference pixel can be generated. The prediction pixel can be generated by linear interpolation of a top interpolation reference pixel and the left interpolation reference pixel or by using a rounded average. Also, in a predetermined number of intra-prediction modes adjacent to mode 6 or 9, a prediction block can be generated using the left reference pixels and the upper reference pixels. In this case, to reduce the complexity, the above-mentioned procedure cannot

ES 2 575 381 T3 be applied in intra-prediction modes having mode numbers greater than a predetermined mode number (eg 9 or 17). Also, the procedure can be applied to only a current prediction unit having a size equal to or greater than a predetermined size. The default size is 8x8 or 16x16.

The prediction block filter 260 adaptively filters the prediction block generated by the prediction block generator 250 according to the intra-prediction mode the current prediction unit received from the prediction mode decoder 230. Prediction block filter 260 may be integrated into prediction block generator 250. A prediction block filtering procedure is the same as that of the prediction block post processor 144 of the intra predictor 140 of FIG. 2.

That is, to reduce the differences in pixel values ​​between a reference pixel and the pixels in the prediction block adjacent to the reference pixel, the prediction block filter 260 adaptively filters out some or all of the pixels in the block. adjacent to the pixel according to the intra-prediction mode. Pixels adjacent to the adjacent pixel exist in the prediction block.

In flat mode, the pixels in the prediction block adjacent to the reference pixel are generated using the reference pixel, and therefore no filter is applied.

In CC mode, an average of reference pixels is used to generate the prediction pixel, and therefore a filter is applied. Different types of filter can be used according to the size of the prediction unit (the size of the prediction block). In a prediction unit of a large size, a filter that is the same as that used in a prediction unit of a small size or a strong filter that has a large uniformity effect can be used.

Image reconstructor 270 receives a prediction block from prediction block generator 250 or prediction block filter 260 in units of prediction units according to the intraprediction mode. Image reconstructor 270 receives a reconstructed residual block by residual signal decoder 220 in units of transform units. Image reconstructor 270 generates a reconstructed image by adding the received prediction block and the residual block. The image can be reconstructed in units of encoding units.

Although the invention has been shown and described with reference to certain exemplary embodiments thereof, it will be understood by those skilled in the art that various changes may be made to the shape and details therein without departing from the scope of the invention. as defined by the appended claims.