JPWO2015045301A1 - Video encoding apparatus, video encoding method, and video encoding program - Google Patents

Abstract

The video encoding device includes a dead zone coefficient selection unit that selects a dead zone coefficient used for truncation of orthogonal transform coefficients based on whether or not the conversion skip mode is valid, and an absolute value from the selected dead zone coefficient. An orthogonal transform coefficient conversion unit that converts an orthogonal transform coefficient that is equal to or less than the derived dead zone to 0, and a quantization unit that quantizes the orthogonal transform coefficient output from the orthogonal transform coefficient conversion unit.

Description

  The present invention relates to a video encoding apparatus to which a dead zone control technique for quantization is applied.

  In the video coding method based on High-Efficiency Video Coding (HEVC) described in Non-Patent Document 1, each frame of digitized video is divided into coding tree units (CTUs) and rasterized. Each CTU is encoded in scan order. Each CTU has a quad-tree structure and is encoded by being divided into coding units (CU: Coding Unit). Each CTU has a quad-tree structure and is encoded by being divided into coding units (CU: Coding Unit). Each CU is predicted by being divided into prediction units (PUs). Further, the prediction error of each CU is divided into transform units (TU: Transform Unit) in a quad-tree structure, and is subjected to frequency conversion.

  CU is a coding unit for intra prediction / interframe prediction. Hereinafter, intra prediction and inter-frame prediction will be described.

  Intra prediction is prediction for generating a predicted image from a reconstructed image of a coding target frame. In Non-Patent Document 1, in addition to DC prediction and Planar prediction, 33 types of angle intra prediction shown in FIG. 12 are defined. In the angle intra prediction, the reconstructed pixels around the encoding target block are extrapolated in any of the 33 types of directions shown in FIG. 12 to generate an intra prediction signal. Hereinafter, a CU that uses intra prediction is referred to as an intra CU.

Inter-frame prediction is prediction in which a prediction image is generated from a reconstructed image of a reconstructed frame (reference picture) having a display time different from that of the encoding target frame. FIG. 13 is an explanatory diagram showing an example of inter-frame prediction. MV = (mv _x , mv _y ) indicates the parallel movement amount of the reconstructed image block in the reference picture with respect to the encoding target block. Inter-frame prediction (hereinafter referred to as inter prediction) generates an inter prediction signal based on the reconstructed image block. Hereinafter, a CU using inter prediction is referred to as an inter CU.

  A frame encoded only by the intra CU is called an I frame (or I picture). A frame including not only an intra CU but also an inter CU is called a P frame (or P picture). A frame that is encoded including not only one reference picture for inter prediction of a block but also an inter CU that uses two reference pictures at the same time is called a B frame (or B picture).

  With reference to FIG. 14, the configuration and operation of a general video encoding apparatus that outputs a bit stream using each CU of each frame of a digitized video as an input image will be described.

  The video encoding apparatus shown in FIG. 14 includes a converter 101, a quantizer 102, an entropy encoder 103, an inverse quantizer / inverse transformer 104, a buffer 105, a predictor 106, and an estimator 107.

  Figure 15 shows an example of CTU division for frame t when the spatial resolution of the frame is CIF (CIF: Common Intermediate Format) and CTU size is 64, and an example of CU division for the eighth CTU (CTU8) included in frame t It is explanatory drawing which shows. FIG. 16 is an explanatory diagram showing a CU quadtree structure of CTUs corresponding to the CU partitioning example of CTU8. It can be seen that the CU quadtree structure is expressed by the CU split flag (cu_split_flag) of each CUDepth.

  FIG. 17 shows a PU divided shape of the CU. If the CU is intra prediction, a square PU partition can be selected (but only 2N × N can be selected if the CU is larger than the minimum size). When CU is inter prediction, if CU is greater than 8, PU partition other than N × N can be selected (however, if CU is 8, only 2N × 2N, 2N × N and N × 2N) Can be selected).

  FIG. 18 is an explanatory diagram illustrating an example of TU partitioning of a CU. The upper part shows an example of TU division of an intra CU having a 2N × 2N PU division shape. When the CU is intra prediction, the root of the quad tree is arranged in the PU, and the TU quad tree structure of the prediction error of the CU is expressed by the TU split flag (tu_split_flag) of each TUDepth. The lower part shows an example of TU partitioning of an inter CU having a 2N × N PU partition shape. When the CU is inter-predicted, the root of the quad tree is arranged in the CU, and the TU quad tree structure of the prediction error of the CU is expressed by the TU split flag (tu_split_flag) of each TUDepth.

  The estimator 107 determines, for each CTU, a CU quadtree structure / PU partition shape / TU quadtree structure that minimizes the coding cost.

  The predictor 106 generates a prediction signal for the input image signal of the CU based on the CU quadtree structure and the PU partition shape determined by the estimator 107. The prediction signal is generated based on the above-described intra prediction or inter prediction.

  Based on the TU quadtree structure determined by the estimator 107 and a transform skip flag (transform_skip_flag) to be described later, the converter 101 performs frequency conversion on a prediction error image obtained by subtracting the prediction signal from the input image signal.

  FIG. 19 is a block diagram illustrating a configuration example of the converter 101. The configuration and operation of converter 101 will be described in more detail with reference to FIG.

  The converter 101 includes a horizontal frequency converter 1011, a first right bit shifter 1012, a vertical frequency converter 1013, a second right bit shifter 1014, a third right bit shifter 1015, and a switch 1010 and a switch 1016. With.

  The converter 101 can use orthogonal transforms of 4 × 4, 8 × 8, 16 × 16, and 32 × 32 block sizes based on frequency transform in transform coding of the prediction error signal. Specifically, integer precision DST (Discrete Sine Transform) can be applied to the luminance component 4 × 4 TU of the intra CU. For other TUs, DCT (Discrete Cosine Transform) with an integer precision corresponding to the block size is applied. FIG. 20 is an explanatory diagram showing the basis of integer precision DCT (integer DCT). FIG. 21 is an explanatory diagram showing a base of integer precision DST (integer DST).

  The base of an integer DCT with an N × N block size (hereinafter also referred to as an N-point integer DCT) is the element of the first N columns of each row every 32 / N rows from the base row of the integer DCT with a 32 × 32 block size. It is obtained by collecting.

  As can be seen from the base value of the integer DCT and the inclusion relationship shown in FIG. 20, the norm of the N-point integer DCT vector is almost uniform for all block sizes. Further, as can be seen from FIG. 21, the base of the integer DST is set so that the norm is substantially uniform with respect to the base of the integer DCT having the same block size.

  The converter 101 can use not only the orthogonal transformation based on the above-described frequency transformation such as DST and DCT but also an orthogonal transformation of 4 × 4 block size based on the unit matrix. This is because frequency conversions such as DST and DCT are not suitable for preserving sharp edges such as screen content and computer graphics (CG). The orthogonal transform based on the unit matrix of 4 × 4 block size is an orthogonal transform that does not perform frequency conversion, and is called a conversion skip mode. Conversion skip mode is enabled when transform_skip_flag is set to 1.

  First, the operation of the converter 101 when transform_skip_flag is 0 will be described.

  The prediction error signal input via the switch 1010 is supplied to the horizontal frequency converter 1011. The horizontal frequency converter 1011 performs frequency conversion of the prediction error signal of the TU having the N × N block size in the horizontal direction using the bases shown in FIG. 20 and FIG.

The first right bit shifter 1012 bit-shifts the prediction error signal of the TU frequency-converted in the horizontal direction to the right by log ₂ N −1 bits.

  Subsequently, the vertical frequency converter 1013 frequency-converts the prediction error signal of the TU supplied from the first right bit shifter 1012 in the vertical direction using the bases shown in FIG. 20 and FIG.

Further, the second right bit shifter 1014 bit-shifts the prediction error signal of the TU supplied from the vertical frequency converter 1013 rightward by log ₂ N +6 bits.

  When transform_skip_flag is 0, the switch 1016 outputs the output of the second right bit shifter 1014 as the orthogonal transform coefficient of the TU.

  Next, the operation of the converter 101 when transform_skip_flag is 1 will be described.

  The prediction error signal of TU of N × N block size input via the switch 1010 is supplied to the third right bit shifter 1015. The third right bit shifter 1015 shifts the prediction error signal to the left by 13 − BitDepth bits. This is because the orthogonal transform coefficient obtained by using the frequency transform and the norm of the orthogonal transform coefficient in the transform skip mode are matched to apply the same quantization. BitDepth is the pixel bit depth of the input image.

  When transform_skip_flag is 1, the switch 1016 outputs the output of the third right bit shifter 1015 as the TU orthogonal transform coefficient.

  The quantizer 102 quantizes the orthogonal transform coefficient supplied from the converter 101. Hereinafter, the quantized orthogonal transform coefficient is referred to as a transform quantization value.

  The operation of the quantizer 102 will be described in more detail.

The quantizer 102 uses linear quantization in which the logarithm of the quantization parameter qP and the quantization step Qstep is proportional (for example, Qstep is doubled when qP is increased by 6). The quantization processing for _obtaining the quantization coefficient q _ij of the orthogonal transformation coefficient c _ij corresponding to the position (i, j) {0 ≦ i, j ≦ N−1} within the TU of the N × N block size is as follows ( 1) It can be defined as

However, Sign () is a function that returns the sign of the input, Int [] is a function that converts the input to an integer value, Qscale is the quantization step coefficient shown in Fig. 22, qP% 6 is the remainder obtained by dividing qP by 6, m _ij is a quantization weighting coefficient, f is an offset (0 ≦ f ≦ 0.5) for determining quantization rounding, and N is a size of TU.

The default value of the quantization weighting coefficient m _ij used for visual image quality adjustment is 16. The quantization weighting coefficient matrix can be transmitted in units of sequences and pictures for each color component, each TU size, and each intra prediction / inter prediction. However, the block size of the quantization weighting coefficient matrix that can be transmitted is up to 8 × 8. The TU quantization weighting coefficient matrix having a block size of 16 × 16 or 32 × 32 is scaled based on a copy of the 8 × 8 quantization weighting coefficient matrix transmitted for each block size TU. However, only the DC component can be transmitted separately.

  FIG. 23, FIG. 24, FIG. 25, and FIG. 26 are explanatory diagrams illustrating default quantization weighting coefficient matrices for intra prediction of 4 × 4, 8 × 8, 16 × 16, and 32 × 32 TU, respectively.

  From FIG. 25, it can be confirmed that the quantization weighting coefficient is the same for each 2 × 2 frequency component in the 16 × 16 intra TU default quantization weighting coefficient matrix by scaling based on the copy. Similarly, from FIG. 26, it can be confirmed that the quantization weighting coefficient is the same for each 4 × 4 frequency component in the 32 × 32 intra TU default quantization weighting coefficient matrix.

  The entropy encoder 103 entropy-encodes cu_split_flag, PU partition shape, tu_split_flag, intra / inter prediction parameter, transform_skip_flag, and transform quantization value.

  The inverse quantization / inverse transformer 104 inversely quantizes the transformed quantized value. Further, the inverse quantizer / inverse transformer 104 inversely transforms the inversely quantized orthogonal transform coefficient. The reconstructed prediction error image that has been inversely transformed is added with a prediction signal and supplied to the buffer 105. The buffer 105 stores the reconstructed image.

  Based on the above-described operation, a general video encoding device generates a bit stream.

"High Efficiency Video Coding (HEVC) text specification draft 10 (for FDIS & Last Call)," JCTVC-L1003_v34, Joint Collaborative Team on Video Coding (JCT-VC) of ITU-T SG 16 WP 3 and ISO / IEC JTC 1 / SC 29 / WG 11 12th Meeting: Geneva, CH, 14-23 Jan. 2013

  The video encoding device described above is the same for the orthogonal transform coefficient of a 4 × 4 block size TU without depending on whether the conversion skip mode is enabled / disabled (whether the conversion skip mode is enabled or disabled). An orthogonal transform coefficient is quantized using a quantization weighting coefficient matrix.

  However, if the same quantization weighting coefficient matrix is used for the prediction error signal without depending on the validity / invalidity of the conversion skip mode, quantization according to visual characteristics cannot be performed.

  FIG. 27 is an explanatory diagram showing a quantization weighting coefficient matrix with a gradient and a uniform quantization weighting coefficient matrix. For example, the quantized weighting coefficient matrix with a slope as shown on the left side in FIG. 27 reduces the amount of information of high frequency components that are difficult to perceive deterioration in a 4 × 4 block size TU in which the conversion skip mode is invalid. Indirectly, low-frequency component quantization noise that easily perceives degradation is reduced. However, the quantization weighting coefficient matrix causes an unacceptable deterioration according to the pixel position in the 4 × 4 block size TU in which the conversion skip mode is effective.

  On the other hand, with respect to a uniform quantization weighting coefficient matrix as shown on the right side in FIG. 27, it is possible to prevent unacceptable deterioration in accordance with the pixel position in the 4 × 4 block size TU in which the conversion skip mode is effective. However, since the quantization weighting coefficient matrix cannot reduce the amount of high-frequency component information in a 4 × 4 block-size TU that is disabled in the conversion skip mode, the quantization of the low-frequency component that easily recognizes deterioration indirectly Increase noise.

  As described above, the technique using the same quantization weighting coefficient matrix for the prediction error signal without depending on the validity / invalidity of the conversion skip mode cannot perform the quantization according to the visual characteristic, and causes local image quality degradation. There is a problem that cannot be avoided.

  In general, a large block size DCT is effective in concentrating power of a prediction error signal in a flat region with little change in sharp pixel values. In addition, the orthogonal transform coefficient by DCT has a characteristic that quantization noise is less perceivable as the frequency component becomes higher.

  Therefore, in the quantization of DCT orthogonal transform coefficients having a block size of 16 × 16 or 32 × 32, it is preferable to use a quantization step size that matches the visual characteristics of each frequency component.

  The video encoding apparatus described above cannot use a quantization step size having a value that matches the visual characteristics of each frequency component in 16 × 16 and 32 × 32 DCT quantization. This is because the quantization weighting coefficient for determining the quantization step size is shared among a plurality of frequency components in order to save the memory capacity on the video decoding device side. For example, even if an attempt is made to apply the quantization weighting coefficient matrix shown in FIG. 28 to an orthogonal transform coefficient of 16 × 16 intra TU, the underlined frequency components in FIG. Quantization cannot be performed with a desired quantization step size.

  FIG. 28 shows a 16 × 16 intra TU quantization weighting coefficient matrix, but the values of the underlined elements are different from the default values shown in FIG.

  When the same quantization weighting coefficient is used for orthogonal transform coefficients of different frequency components, it is not possible to efficiently remove frequency components with small power that are difficult to perceive quantization noise. For example, since there is noise due to analog-to-digital conversion in the video, a high frequency component with a small power is generated even in a flat region where a 16 × 16 or 32 × 32 DCT is selected. Since the remaining frequency components increase entropy, in order to keep the bit rate low, the orthogonal transform coefficients of the low frequency components must be roughly quantized. As a result, the above-described video encoding device increases the low-frequency component quantization noise and reduces the video quality.

  It is an object of the present invention to provide a video encoding device, a video encoding method, and a video encoding program that can suppress a decrease in video quality.

  The video encoding apparatus according to the present invention includes dead zone coefficient selection means for selecting a dead zone coefficient used for truncation of orthogonal transform coefficients based on whether or not the transform skip mode is valid, and a dead value whose absolute value is selected. An orthogonal transform coefficient converting unit that converts an orthogonal transform coefficient that is equal to or less than a dead zone derived from a zone coefficient to 0, and a quantizing unit that quantizes the orthogonal transform coefficient output by the orthogonal transform coefficient converting unit. And

  Another video encoding device according to the present invention is derived from dead zone coefficient setting means for setting a dead zone coefficient corresponding to each frequency component of an orthogonal transform coefficient having a block size larger than a predetermined size, and the set dead zone coefficient. Characterized in that it comprises orthogonal transform coefficient transform means for transforming an orthogonal transform coefficient of an absolute value that is equal to or less than a dead zone to 0, and quantization means for quantizing the orthogonal transform coefficient output by the orthogonal transform coefficient transform means. To do.

  The video coding method according to the present invention selects a dead zone coefficient to be used for truncation of orthogonal transform coefficients based on whether or not the transform skip mode is valid, and an absolute value is derived from the selected dead zone coefficients. An orthogonal transform coefficient conversion process for converting an orthogonal transform coefficient that is equal to or less than a dead zone to 0 is executed, and the orthogonal transform coefficient by the orthogonal transform coefficient conversion process is quantized.

  Another video encoding method according to the present invention sets a dead zone coefficient corresponding to each frequency component of an orthogonal transform coefficient having a block size larger than a predetermined size, and is equal to or less than a dead zone derived from the set dead zone coefficient. An orthogonal transform coefficient transform process for transforming an orthogonal transform coefficient of a certain absolute value to 0 is executed, and the orthogonal transform coefficient by the orthogonal transform coefficient transform process is quantized.

  According to the video encoding program of the present invention, a computer selects a dead zone coefficient used for truncation of orthogonal transform coefficients based on whether or not the transform skip mode is valid, and an absolute value of the selected dead zone. An orthogonal transform coefficient transform process for transforming an orthogonal transform coefficient that is equal to or less than a dead zone derived from the coefficients to 0 and a process for quantizing the orthogonal transform coefficient by the orthogonal transform coefficient transform process are executed.

  Another video encoding program according to the present invention is derived from a process for setting a dead zone coefficient corresponding to each frequency component of an orthogonal transform coefficient having a block size larger than a predetermined size in the computer, and the set dead zone coefficient. An orthogonal transform coefficient transform process for transforming an orthogonal transform coefficient with an absolute value that is equal to or less than a dead zone to 0 and a process for quantizing the orthogonal transform coefficient by the orthogonal transform coefficient transform process are executed.

  According to the present invention, it is possible to suppress a decrease in video quality.

It is a block diagram which shows the structure of one Embodiment of a video coding apparatus. It is a block diagram which shows the structural example of an adaptive truncator. It is explanatory drawing which shows a dead zone coefficient matrix and a quantization weighting coefficient matrix. It is a flowchart which shows operation | movement of an adaptive truncator. It is explanatory drawing which shows a non-uniform dead zone coefficient matrix. It is a flowchart which shows operation | movement of an adaptive truncator. It is explanatory drawing which shows the 2x2 sub dead zone coefficient matrix used for a 16x16 intra TU. It is explanatory drawing which shows the 4x4 sub dead zone coefficient matrix used for a 32x32 intra TU. It is a block diagram which shows the structural example of the information processing system which can implement | achieve the function of a video coding apparatus. It is a block diagram which shows the principal part of the video coding apparatus by this invention. It is a block diagram which shows the principal part of the other video coding apparatus by this invention. It is explanatory drawing which shows the example of 33 types of angle intra prediction. It is explanatory drawing which shows the example of inter-frame prediction. It is explanatory drawing which shows the structure of a general video coding apparatus. FIG. 6 is an explanatory diagram showing an example of CTU partitioning for frame t and an example of CU partitioning for CTU8 in frame t. It is explanatory drawing which shows the quadtree structure corresponding to the CU division | segmentation example of CTU8. It is explanatory drawing which shows the example of PU division of CU. It is explanatory drawing which shows the TU division | segmentation example of CU. It is a block diagram which shows the structural example of a converter. It is explanatory drawing which shows the basis of DCT of integer precision. It is explanatory drawing which shows the base of DST of integer precision. It is explanatory drawing which shows a quantization step coefficient. It is explanatory drawing which shows the default quantization weighting coefficient matrix of 4 * 4TU intra prediction. It is explanatory drawing which shows the default quantization weighting coefficient matrix of 8 * 8TU intra prediction. It is explanatory drawing which shows the default quantization weighting coefficient matrix of 16 * 16TU intra prediction. It is explanatory drawing which shows the default quantization weighting coefficient matrix of 32 * 32TU intra prediction. It is explanatory drawing which shows the quantization weighting coefficient matrix with a gradient, and a uniform quantization weighting coefficient matrix. It is explanatory drawing which shows the quantization weighting coefficient matrix with a gradient, and a uniform quantization weighting coefficient matrix.

Embodiment 1. FIG.
FIG. 1 is a block diagram illustrating a configuration of an embodiment of a video encoding device. The video encoding apparatus shown in FIG. 1 is adaptive in addition to a converter 101, a quantizer 102, an entropy encoder 103, an inverse quantizer / inverse transformer 104, a buffer 105, a predictor 106, and an estimator 107. A truncator 108 is provided.

  The estimator 107 determines a CU quadtree structure / PU partition shape / TU quadtree structure that minimizes the coding cost for each CTU.

  The predictor 106 generates a prediction signal for the input image signal of the CU based on the CU quadtree structure and the PU partition shape determined by the estimator 107.

  Based on the TU quadtree structure and the transform skip flag (transform_skip_flag) determined by the estimator 107, the converter 101 performs frequency conversion on the prediction error image obtained by subtracting the prediction signal from the input image signal.

  The adaptive truncator 108 performs a truncation process on the orthogonal transform coefficient supplied from the transformer 101 based on the transform skip flag (transform_skip_flag) determined by the estimator 107.

  FIG. 2 is a block diagram illustrating a configuration example of the adaptive truncator 108. The adaptive truncator 108 uses the dead zone coefficient matrix and the quantization weighting coefficient matrix shown in FIG. 3 for the orthogonal transform coefficients of 4 × 4 block size TU obtained by the intra prediction. FIG. 3 is an explanatory diagram showing a dead zone coefficient matrix and a quantization weighting coefficient matrix used by the video encoding apparatus of the present embodiment.

  The dead zone technique is a technique for setting an output corresponding to an input close to 0 to 0 when quantizing orthogonal transform coefficients. The dead zone corresponds to an index for determining whether or not to set the orthogonal transform coefficient to zero. The dead zone coefficient is a parameter used for setting the dead zone, and can be arbitrarily set.

  The sloped dead zone coefficient matrix shown on the left side in Fig. 3 is obtained by dividing each element of the sloped quantization weighting coefficient matrix shown on the left side in Fig. 27 by the element of the uniform quantization weighting coefficient matrix shown in the lower center of Fig. 3. The matrix generated by It can be seen that the dead zone coefficient increases as i and j indicating the position (i, j) {0 ≦ i, j ≦ 3} in the TU increase. The uniform dead zone coefficient matrix shown on the right side in FIG. 3 is obtained by dividing each element of the uniform quantization weight coefficient matrix shown on the right side in FIG. 27 by the element of the uniform quantization weight coefficient matrix shown in the lower center of FIG. This is a generated matrix. It can be seen that the dead zone coefficient has a fixed value without depending on i and j indicating the position (i, j) {0 ≦ i, j ≦ 3} in the TU.

When the transform_skip_flag is 1, the dead zone coefficient selection unit 1081 selects the above-described uniform dead zone coefficient matrix. When transform_skip_flag is 0, the above-described sloped dead zone coefficient matrix is selected. Hereinafter, let the dead zone coefficient corresponding to the position (i, j) be _dij .

Dead zone calculation unit 1082 calculates the dead zone dz _ij for orthogonal transformation coefficients c _ij by the following equation (2).

From the dead zone coefficient matrix shown in FIG. 3, the dead zone dz _ij calculated by the above formula has the following characteristics.

When transform_skip_flag is 0, i and j indicating the position (i, j) {0 ≦ i, j ≦ 3} in TU become larger (that is, the higher the frequency component, the higher the frequency component), the more the dead zone dz _ij becomes growing. That is, the amount of information on orthogonal transform coefficients of high-frequency components that are less likely to deteriorate can be reduced by the processing of the truncation unit 1083 described later.

When transform_skip_flag is 1, the dead zone dz _ij becomes a fixed value without depending on i and j indicating the position (i, j) {0 ≦ i, j ≦ 3} in the TU. That is, the amount of information of the orthogonal transform coefficient can be reduced by a constant amount without depending on the pixel position by the processing of the truncation unit 1083 described later.

The truncation unit 1083 compares the absolute value of c _ij at each position (i, j) {0 ≦ i, j ≦ 3} with dz _ij only for the TU orthogonal transform coefficient of 4 × 4 block size, When the absolute value of c _ij is less than dz _ij , c _ij is output as 0. In other cases, c _ij is output as it is. The following formula (3) is typical.

  In this embodiment, the quantizer 102 quantizes the orthogonal transform coefficient supplied from the adaptive truncator 108.

  The entropy encoder 103 entropy-encodes cu_split_flag, PU partition shape, tu_split_flag, intra / inter prediction parameter, transform_skip_flag, and transform quantization value.

  The inverse quantization / inverse transformer 104 inversely quantizes the transformed quantized value. Further, the inverse quantizer / inverse transformer 104 inversely transforms the inversely quantized orthogonal transform coefficient. The reconstructed prediction error image that has been inversely transformed is added with a prediction signal and supplied to the buffer 105. The buffer 105 stores the reconstructed image.

  Based on the above-described operation, the video encoding apparatus according to the present embodiment generates a bit stream.

  Next, the operation of the adaptive truncator 108 for a 4 × 4 block size TU will be described with reference to FIG.

  In step S101, the dead zone coefficient selection unit 1081 determines whether or not the transform_skip_flag of the 4 × 4 block size TU to be processed is 0. If it is 0, the process proceeds to step S102. Otherwise, the process proceeds to step S103.

  In step S102, the dead zone coefficient selection unit 1081 selects the sloped dead zone coefficient matrix shown on the left side in FIG. Then, the process proceeds to step S104.

  In step S103, the dead zone coefficient selection unit 1081 selects the uniform dead zone coefficient matrix shown on the right side in FIG. Then, the process proceeds to step S104.

In step S104, the dead zone calculation unit 1082 determines, based on the selected dead zone coefficient matrix, each position (i, j) {0 ≦ i, j ≦ 3} of the 4 × 4 block size TU to be processed. Compute dead zone dz _ij for orthogonal transform coefficients.

In step S105, the truncation unit 1083 determines the absolute value of the orthogonal transform coefficient c _ij at each position (i, j) {0 ≦ i, j ≦ 3} of the 4 × 4 block size TU to be processed and the dead zone dz _ij It compares the absolute value of c _ij are the following cases dz _ij, and outputs the c _ij 0. In other cases, truncation section 1083 outputs c _ij as it is.

  As described above, the adaptive truncator 108 monitors the validity / invalidity of the transform skip mode (transform_skip_flag value) of the processing target block and uses the dead zone coefficient used for truncation of the orthogonal transform coefficient of the processing target 4 × 4 block size TU. Is selected adaptively. Specifically, when the conversion skip mode of 4 × 4 block size TU is invalid, a dead zone coefficient with a slope is selected. When the 4 × 4 block size TU conversion skip mode is enabled, a uniform dead zone coefficient is selected. Further, the adaptive truncator 108 calculates a dead zone from the selected dead zone coefficient and quantization step size, and converts the orthogonal transform coefficient included in the 4 × 4 block size TU whose absolute value is equal to or smaller than the dead zone to 0. .

  Since the video encoding device of the present embodiment includes the adaptive truncator 108, even if a video encoding scheme using a uniform quantization weighting coefficient matrix is used, the 4 × 4 block in which the conversion skip mode is disabled is selected. In the size TU, it is possible to reduce the amount of information of high frequency components that are difficult to perceive degradation in a dead zone with a slope. As a result, it is possible to indirectly reduce low-frequency component quantization noise that easily perceives deterioration. Therefore, even if the video encoding method uses the same uniform quantization weighting coefficient matrix for orthogonal transform coefficients of 4 × 4 block size TU without depending on the validity / invalidity of the transform skip mode, the visual characteristics It is possible to avoid local image quality degradation by quantization according to.

  In this embodiment, the dead zone coefficient matrix and the quantization weighting coefficient matrix shown in FIG. 3 are used for the 4 × 4 block size TU orthogonal transform coefficient obtained by intra prediction, but it is obtained by inter prediction. Similarly, for 4 × 4 block size TU orthogonal transform coefficients, in the 4 × 4 block size TU for which the conversion skip mode is disabled, the amount of information of high-frequency components that are difficult to perceive deterioration is slanted. Can be reduced in the zone.

In the embodiment described above, the elements of the sloped dead zone coefficient matrix and the uniform dead zone coefficient matrix are stored in a predetermined storage unit as real numbers. However, using the fact that the quantization step Qstep doubles when the quantization parameter qP increases by 6 in quantization, the elements of the sloped dead zone coefficient matrix and uniform dead zone coefficient matrix are stored as integers in the storage unit. It is also possible to do. The real dead zone coefficient dz _ij can be expressed as an integer dead zone coefficient qPDZOffset _ij by the logarithmic expression of the following equation.

qPDZOffset _ij = 6 ・ log ₂ dz _ij

When the dead zone coefficient qPDZOffset _ij is used, the dead zone calculation unit 1082 calculates the dead zone dz _ij for the orthogonal transformation coefficient c _{ij as} shown in the following equation (4).

As can be seen from the above qPDZ _ij calculation formula, it can be seen that the dead zone coefficient qPDZOffset _ij is used as an offset of the quantization parameter qP. By using integer dead zone coefficients instead of real numbers, the storage capacity of the sloped dead zone coefficient matrix and the uniform dead zone coefficient matrix can be reduced.

  The video encoding apparatus according to the present embodiment adaptively removes high-frequency component orthogonal transform coefficients that are invalid in the transform skip mode when using a video coding system that uses a uniform quantization weighting coefficient matrix. Quantization according to visual characteristics can be realized, and local image quality degradation can be avoided.

  That is, the video encoding apparatus of the present embodiment does not deteriorate even in the 4 × 4 block size TU in which the conversion skip mode invalidity is selected even when using a video encoding scheme that uses a uniform quantization weighting coefficient matrix. The amount of information of high frequency components that are difficult to recognize can be reduced in a dead zone with a slope. As a result, it is possible to indirectly reduce low-frequency component quantization noise that easily perceives deterioration. Therefore, even if the video encoding method uses the same uniform quantization weighting coefficient matrix for orthogonal transform coefficients of 4 × 4 block size TU without depending on the validity / invalidity of the transform skip mode, the visual characteristics By the quantization according to the local image quality degradation is avoided.

Embodiment 2. FIG.
Next, a second embodiment of the video encoding device will be described. The configuration of the video encoding apparatus according to the second embodiment is as shown in FIG.

  Hereinafter, in order to simplify the description, an example is given in which the converter 101 uses orthogonal transform based on DCT of 4 × 4, 8 × 8, and 16 × 16 block sizes.

  In the second embodiment, the adaptive truncator 108 shown in FIG. 2 uses the non-uniform dead zone coefficient matrix shown in FIG. 5 for the orthogonal transform coefficient of 16 × 16 block size TU obtained by intra prediction. And the quantization weighting coefficient matrix shown in FIG. 25 is used as the quantization weighting matrix to be transmitted.

  The dead zone technique is a technique for setting an output corresponding to an input close to 0 to 0 when quantizing orthogonal transform coefficients. The dead zone corresponds to an index for determining whether or not to set the orthogonal transform coefficient to zero. The dead zone coefficient is a parameter used for setting the dead zone, and can be arbitrarily set.

  The non-uniform dead zone coefficient matrix shown in FIG. 5 is a matrix generated by dividing each element of the quantization weighting coefficient matrix shown in FIG. 28 by the element of the default quantization weighting coefficient matrix shown in FIG. It can be seen that the dead zone coefficient varies depending on the position (i, j) {0 ≦ i, j ≦ N−1} in the TU, that is, depending on the frequency component.

The dead zone coefficient selection unit 1081 selects the non-uniform dead zone coefficient matrix described above when the DCT score N is larger than 8. When the DCT score N is 8 or less, the above-described uniform dead zone coefficient matrix is selected. Let d _{ij be the} dead zone coefficient corresponding to the frequency component (i, j).

Dead zone calculation unit 1082, when DCT of points N is larger than 8, the dead zone dz _ij for orthogonal transformation coefficients c _ij are calculated as above in (2) below.

From the dead zone coefficient matrix shown in FIG. 5, the dead zone dz _ij calculated by equation (2) has the following characteristics.

Depending on the frequency component (i, j) {0 ≦ i, j ≦ N−1} in the TU, the dead zone dz _ij takes different values. In other words, by the processing of the truncation unit 1083 described later, the amount of information on the orthogonal transform coefficient of the frequency component that is less likely to be deteriorated is reduced even if the orthogonal transform coefficient is the same quantization weighting coefficient every 2 × 2. Can do.

The truncation unit 1083 applies c _{ij of} each frequency component (i, j) {0 ≦ i, j ≦ N−1} to the TU orthogonal transform coefficient of 16 × 16 block size larger than 8 × 8 block size. of comparing the absolute value and dz _ij, the absolute value of c _ij are the following cases dz _ij, and outputs the c _ij 0. In other cases, c _ij is output as it is. The above formula (3) is typical.

  In this embodiment, the quantizer 102 quantizes the orthogonal transform coefficient supplied from the adaptive truncator 108.

  The entropy encoder 103 entropy-encodes cu_split_flag, PU partition shape, tu_split_flag, intra / inter prediction parameter, transform_skip_flag, and transform quantization value.

  The inverse quantization / inverse transformer 104 inversely quantizes the transformed quantized value. Further, the inverse quantizer / inverse transformer 104 inversely transforms the inversely quantized orthogonal transform coefficient. The reconstructed prediction error image that has been inversely transformed is added with a prediction signal and supplied to the buffer 105. The buffer 105 stores the reconstructed image.

  Based on the above-described operation, the video encoding apparatus according to the present embodiment generates a bit stream.

  Next, the operation of the adaptive truncator 108 for a 4 × 4 block size TU will be described with reference to FIG.

  In step S201, the dead zone coefficient selection unit 1081 determines whether or not the block size of the DCT to be processed is larger than 8 × 8. If larger than 8 × 8, that is, if the block size of the DCT to be processed is 16 × 16, the process proceeds to step S202. Otherwise, the process ends.

  In step S202, the dead zone coefficient selection unit 1081 selects the non-uniform dead zone coefficient matrix shown in FIG. Then, the process proceeds to step S203.

  Note that the dead zone coefficient selection unit 1081 selects a uniform dead zone coefficient in which all elements are 1 when the block size of the DCT is 8 × 8 or less.

In step S203, the dead zone calculation unit 1082 determines, based on the selected dead zone coefficient matrix, each frequency component (i, j) {0 ≦ i, j ≦ 15} of the 16 × 16 block size TU to be processed Calculate the dead zone dz _ij for the orthogonal transform coefficient of. Then, the process proceeds to step S204.

In step S204, the truncation unit 1083 calculates the absolute value of the orthogonal transform coefficient c _ij at each position (i, j) {0 ≦ i, j ≦ 15} of the 16 × 16 block size TU to be processed and the dead zone dz _ij It compares the absolute value of c _ij are the following cases dz _ij, and outputs the c _ij 0. In other cases, truncation section 1083 outputs c _ij as it is.

  The quantizer 102 quantizes the orthogonal transform coefficient output from the adaptive truncator 108. When the block size is larger than 8 × 8, a plurality of frequency components are used for the orthogonal transform coefficient (excluding the DC component). Quantize orthogonal transform coefficients using quantization weighting coefficients of the same value for. For example, the quantization weighting coefficient matrix shown in FIG. 23 or FIG. 24 is applied to the orthogonal transform coefficient.

  As described above, the adaptive truncator 108 monitors the DCT block size of the processing target block and adaptively selects the dead zone used for truncation of the orthogonal transform coefficient of the processing target block. Specifically, when the DCT block size of the processing target block is larger than 8 × 8 (that is, the same quantization weighting coefficient is used for orthogonal transform coefficients of frequency components having different block sizes of the processing target block). Non-uniform dead zone coefficients that can take different values for each frequency component. Otherwise, select a uniform dead zone factor with a fixed value of 1. Further, the adaptive truncator 108 calculates a dead zone from the selected dead zone coefficient and quantization step size, and converts the orthogonal transform coefficient of a small high frequency component whose absolute value is equal to or less than the dead zone to 0.

  Since the video encoding apparatus of the present embodiment includes the adaptive truncator 108, even if a 16 × 16 block size DCT that uses the same quantization weighting coefficient for quantization for each 2 × 2 frequency component is used, the frequency A frequency component with a small power can be efficiently removed according to the visual characteristics of each component. As a result, low-frequency component quantization noise is reduced, and video quality can be improved. That is, by suppressing an increase in quantization noise of low frequency components, it is possible to suppress a decrease in video quality.

  In this embodiment, to simplify the description, the converter 101 uses orthogonal transform based on DCT of 4 × 4, 8 × 8, and 16 × 16 block sizes. Even if a 32 × 32 block size DCT using the same quantization weighting coefficient for each frequency component of × 4 is used, as in the case of the 16 × 16 block size, the adaptive truncator 108 A frequency component with a small power can be efficiently removed according to the visual characteristics.

  The adaptive truncator 108 is also used for the 16 × 16 block size obtained by inter prediction and the 32 × 32 block size TU orthogonal transform coefficient, similarly to the 16 × 16 block size obtained by intra prediction. Thus, frequency components with small power can be efficiently removed according to visual characteristics for each frequency component.

  In the embodiment described above, all elements of the 16 × 16 block size dead zone coefficient matrix are stored in the truncation unit 1083. Focusing on the fact that 2 × 2 frequency components have the same quantization weighting coefficient, a 2 × 2 sub-dead zone coefficient matrix shown in FIG. 7 may be used for 16 × 16 intra TUs.

  In the embodiment described above, all elements of the 16 × 16 block size dead zone coefficient matrix are stored in the truncation unit 1083. However, a 2 × 2 sub-dead zone coefficient matrix shown in FIG. 7 may be used by paying attention to the fact that 2 × 2 frequency components have the same quantization weighting coefficient.

In this case, the dead zone dz _ij (0 ≦ i, j ≦ 15) of 16 × 16 block size is calculated by the following equation (5).

  Thus, by using the sub dead zone coefficient matrix corresponding to copying, the storage capacity of the dead zone coefficient matrix can be reduced. Similarly, for a DCT of 32 × 32 block size, paying attention to the same quantization weighting factor for each 4 × 4 frequency component, the 4 × 4 shown in FIG. A sub-dead zone coefficient matrix of 4 may be used.

In this case, the dead zone dz _ij (0 ≦ i, j ≦ 31) of 32 × 32 block size is calculated by the following equation (6).

In the embodiment described above, the elements of the dead zone coefficient matrix are stored as real numbers in a predetermined storage unit. However, by utilizing the fact that the quantization step Qstep doubles when the quantization parameter qP increases by 6 in quantization, the elements of the dead zone coefficient matrix can be stored in the storage unit as integers. The real dead zone coefficient dz _ij can be expressed as an integer dead zone coefficient qPDZOffset _ij by the logarithmic expression of the following equation.

qPDZOffset _ij = 6 ・ log ₂ dz _ij

When the dead zone coefficient qPDZOffset _ij is used, the dead zone calculation unit 1082 calculates the dead zone dz _ij for the orthogonal transformation coefficient c _{ij as} shown in the above equation (4).

As can be seen from the above qPDZ _ij calculation formula, it can be seen that the dead zone coefficient qPDZOffset _ij is used as an offset of the quantization parameter qP. By using an integer dead zone coefficient instead of a real number, the storage capacity of the dead zone coefficient matrix can be further reduced.

  Moreover, although each said embodiment can also be comprised with a hardware, it is also possible to implement | achieve by a computer program.

  The information processing system shown in FIG. 9 includes a processor 1001, a program memory 1002, a storage medium 1003 for storing video data, and a storage medium 1004 for storing a bitstream. The storage medium 1003 and the storage medium 1004 may be separate storage media, or may be storage areas composed of the same storage medium. A magnetic storage medium such as a hard disk can be used as the storage medium.

  In the information processing system shown in FIG. 9, the program memory 1002 stores a program for realizing the function of each block shown in FIG. The processor 1001 implements the function of the video encoding device shown in FIG. 1 by executing processing according to the program stored in the program memory 1002.

  FIG. 10 is a block diagram showing a main part of a video encoding device according to the present invention. As shown in FIG. 10, the video encoding apparatus includes a dead zone coefficient selection unit 10 (for example, shown in FIG. 2) that selects a dead zone coefficient used for truncation of orthogonal transform coefficients based on the validity / invalidity of the transform skip mode. A dead zone coefficient selecting unit 1081) and an orthogonal transform coefficient converting unit 20 that converts an orthogonal transform coefficient whose absolute value is equal to or less than the dead zone derived from the selected dead zone coefficient to 0 (as shown in FIG. 2 as an example) A dead zone calculation unit 1082 and a truncation unit 1083), and a quantization unit 30 (quantizer 102 shown in FIG. 1 as an example) that quantizes the orthogonal transformation coefficient output from the orthogonal transformation coefficient transformation unit 20.

  FIG. 11 is a block diagram showing a main part of another video encoding apparatus according to the present invention. As shown in FIG. 11, the video encoding apparatus includes a dead zone coefficient setting unit 11 that sets a dead zone coefficient corresponding to each frequency component of an orthogonal transform coefficient having a block size larger than a predetermined size (for example, FIG. Dead zone coefficient selection unit 1081) and an orthogonal transform coefficient conversion unit 21 that converts the orthogonal transform coefficient of the absolute value whose absolute value is equal to or less than the dead zone derived from the set dead zone coefficient to 0 (as an example, 2, a dead zone calculation unit 1082 and a truncation unit 1083), a quantization unit 31 (for example, the quantizer 102 shown in FIG. 1) that quantizes the orthogonal transformation coefficient output from the orthogonal transformation coefficient transformation unit 21, and Is provided.

  While the present invention has been described with reference to the embodiments and examples, the present invention is not limited to the above embodiments and examples. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  This application claims priority based on Japanese Patent Application No. 2013-200892 filed on Sep. 27, 2013 and Japanese Patent Application No. 2013-204864 filed on Sep. 30, 2013. Get everything here.

10 Dead zone coefficient selector
11 Dead zone coefficient setting section
20,21 Orthogonal transform coefficient transform unit
30,31 Quantizer
101 converter
102 Quantizer
103 Entropy encoder
104 Inverse quantization / inverse transformer
105 buffers
106 Predictor
107 Estimator
108 Adaptive truncator
1081 Dead zone coefficient selector
1082 Dead zone calculator
1083 truncation part

Claims (20)

Dead zone coefficient selection means for selecting a dead zone coefficient to be used for truncation of orthogonal transform coefficients based on whether or not the transform skip mode is valid;
Orthogonal transform coefficient transforming means for transforming an orthogonal transform coefficient whose absolute value is equal to or less than the dead zone derived from the selected dead zone coefficient to 0;
A video encoding device comprising: quantization means for quantizing orthogonal transform coefficients output from the orthogonal transform coefficient transform means. The video encoding apparatus according to claim 1, wherein the quantization means quantizes the orthogonal transform coefficient using the same quantization weighting coefficient regardless of whether the transform skip mode is valid or invalid. The video encoding device according to claim 2, wherein the orthogonal transform coefficient conversion means outputs an orthogonal transform coefficient having an absolute value larger than that of the dead zone without being converted. The orthogonal transform coefficient transform unit derives a dead zone that becomes a fixed value when the transform skip mode is invalid and increases as the frequency component becomes higher when the transform skip mode is valid. The video encoding device according to claim 2 or claim 3. A dead zone coefficient setting means for setting a dead zone coefficient corresponding to each frequency component of the orthogonal transform coefficient having a block size larger than a predetermined size;
Orthogonal transform coefficient conversion means for converting an orthogonal transform coefficient of an absolute value equal to or less than a dead zone derived from a set dead zone coefficient to 0;
A video encoding device comprising: quantization means for quantizing orthogonal transform coefficients output from the orthogonal transform coefficient transform means. The video encoding apparatus according to claim 5, wherein the quantization means quantizes the orthogonal transform coefficient by using quantization weighting coefficients having the same value for a plurality of frequency components with respect to the orthogonal transform coefficient. The video encoding device according to claim 6, wherein the orthogonal transform coefficient conversion unit derives a dead zone having a different value according to a frequency component. The video encoding device according to claim 6 or 7, wherein the orthogonal transform coefficient conversion means outputs an orthogonal transform coefficient having an absolute value larger than that of a dead zone without being converted. Select a dead zone coefficient to be used for truncation of orthogonal transform coefficients based on whether transform skip mode is enabled,
Performing an orthogonal transform coefficient transform process for transforming an orthogonal transform coefficient whose absolute value is equal to or less than the dead zone derived from the selected dead zone coefficient to 0;
A video encoding method, wherein the orthogonal transform coefficient obtained by the orthogonal transform coefficient transform process is quantized. The video encoding method according to claim 9, wherein the orthogonal transform coefficient is quantized using the same quantization weighting coefficient regardless of whether the transform skip mode is valid or invalid. The video according to claim 9 or 10, wherein a dead zone is derived when the conversion skip mode is invalid and becomes a fixed value, and when the conversion skip mode is valid, the value increases as the frequency component becomes higher. Encoding method. Set the dead zone coefficient corresponding to each frequency component of the orthogonal transform coefficient of the block size larger than the predetermined size,
An orthogonal transform coefficient conversion process for converting an orthogonal transform coefficient of an absolute value that is equal to or less than a dead zone derived from a set dead zone coefficient to 0 is executed,
A video encoding method, wherein the orthogonal transform coefficient obtained by the orthogonal transform coefficient transform process is quantized. The video encoding method according to claim 12, wherein the orthogonal transform coefficient is quantized with respect to the orthogonal transform coefficient by using quantization weighting coefficients having the same value for a plurality of frequency components. The video encoding method according to claim 12 or 13, wherein dead zones having different values according to frequency components are derived. On the computer,
A process of selecting a dead zone coefficient to be used for truncation of the orthogonal transform coefficient based on whether the transform skip mode is valid;
An orthogonal transform coefficient transform process for transforming an orthogonal transform coefficient whose absolute value is equal to or less than a dead zone derived from the selected dead zone coefficient to 0;
A video encoding program for executing a process of quantizing an orthogonal transform coefficient by the orthogonal transform coefficient conversion process. On the computer,
The video coding program according to claim 15, wherein the orthogonal transform coefficient is quantized using the same quantization weighting coefficient regardless of whether the transform skip mode is valid or invalid. On the computer,
The video according to claim 15 or 16, wherein a dead zone is derived when the conversion skip mode is invalid and becomes a fixed value, and when the conversion skip mode is valid, the value increases as the frequency component becomes higher. Encoding program. On the computer,
A process of setting a dead zone coefficient corresponding to each frequency component of an orthogonal transform coefficient having a block size larger than a predetermined size;
An orthogonal transform coefficient conversion process for converting an orthogonal transform coefficient of an absolute value that is equal to or less than a dead zone derived from a set dead zone coefficient to 0;
A video encoding program for executing a process of quantizing an orthogonal transform coefficient by the orthogonal transform coefficient conversion process. On the computer,
19. The video encoding program according to claim 18, wherein the orthogonal transform coefficient is quantized with respect to the orthogonal transform coefficient by using quantization weighting coefficients having the same value for a plurality of frequency components. On the computer,
The video encoding program according to claim 18 or 19, wherein a dead zone having a different value according to a frequency component is derived.

Images