JP5108828B2 - Image encoding device - Google Patents

Description

  The present invention relates to an image encoding device and an image encoding method.

  When processing such as digitizing and storing a moving image is performed, the amount of data becomes enormous, and thus the moving image is compressed. By storing the compressed data in a hard disk or the like, it is possible to store moving image data for a long time. As such an image compression method, a method of using a correlation between adjacent blocks in a frame to increase the compression rate has been studied. For example, the moving picture coding system H.264 In H.264, intra prediction coding using correlation between blocks is employed.

  FIG. 26 shows a block configuration diagram of the image encoding device 1. As shown in FIG. 26, the image encoding device 1 includes a blocking unit 2, a difference calculation unit 3, a DCT unit 4, a quantization unit 5, an encoding unit 6, an inverse quantization unit 7, An inverse DCT unit 8, a reconstruction unit 9, an adjacent pixel storage unit 10, a prediction mode determination unit 11, and an intra prediction unit 12 are included.

  The blocking unit 2 blocks the input image data into a predetermined block unit (color difference signal: 8 pixels × 8 pixels, luminance signal 16 pixels × 16 pixels) for the luminance signal and the color difference signal.

  The intra prediction unit 12 generates prediction data from adjacent pixels accumulated in the adjacent pixel storage unit 10 in the prediction mode determined by the prediction mode determination unit 11.

  The difference calculation unit 3 calculates the difference between the predicted pixel data generated by the intra prediction unit 12 and the pixels of the input image data. The DCT unit 4 performs DCT conversion processing on the difference data from the difference calculation unit 3. The quantization unit 5 quantizes the data after the DCT conversion of the DCT unit 4.

  The encoding unit 6 encodes the output data of the quantization unit 5. The inverse quantization unit 7 performs inverse quantization on the output data of the quantization unit 5. The inverse DCT unit 8 performs an inverse DCT transform process on the data inversely quantized by the inverse quantization unit 7. The reconstruction unit 9 reconstructs image data from the output data from the inverse DCT unit 8 and the prediction data from the intra prediction unit 12. The adjacent pixel storage unit 10 stores adjacent pixel data at the boundary of each block from the output data from the reconstruction unit 9.

  The prediction mode determination unit 11 generates a prediction pixel predicted from the adjacent pixel data stored in the adjacent pixel storage unit 10. Then, for each prediction mode, the difference between the predicted pixel value and the input pixel value is calculated, and the prediction mode with the smallest difference value is determined as the optimum prediction mode. Then, the intra prediction unit 12 again generates a prediction pixel in the determined optimal prediction mode, and the difference calculation unit 3 calculates the difference between the prediction pixel and the input pixel.

  As an example of the prediction mode determination unit 11, there is a technique as described in Non-Patent Document 1. FIG. 27 shows a schematic block diagram of Non-Patent Document 1. Here, it is assumed that DC prediction, vertical prediction, horizontal prediction, and plane prediction are adopted as the types of prediction modes.

  As illustrated in FIG. 27, the prediction mode determination unit 11 of Non-Patent Document 1 includes a DC prediction pixel generation unit 21, a vertical prediction pixel generation unit 22, a horizontal prediction pixel generation unit 23, a planar prediction pixel generation unit 24, and a difference calculation. Units 31 to 34, two-dimensional Amadal transformation value generation units 41 to 44, absolute value sum calculation units 51 to 54, and comparison unit 60.

  The DC prediction pixel generation unit 21, the vertical prediction pixel generation unit 22, the horizontal prediction pixel generation unit 23, and the planar prediction pixel generation unit 24 generate prediction pixels for each prediction mode according to adjacent reference pixels adjacent to the block. .

  The difference calculation units 31 to 34 receive the pixels stored in the adjacent pixel storage unit 10 of FIG. 26 (hereinafter referred to as input pixels), and the DC prediction pixel generation unit 21, the vertical prediction pixel generation unit 22, and the horizontal prediction. The difference with each prediction pixel which the pixel production | generation part 23 and the planar prediction pixel production | generation part 24 produced | generated is calculated.

  The two-dimensional amadar conversion value generation units 41 to 44 perform two-dimensional amadar conversion processing on the difference values output from the difference calculation units 31 to 34, respectively. Then, the absolute value sum calculation units 51 to 54 calculate the absolute value sums of the conversion values processed by the two-dimensional Amadal conversion value generation units 41 to 44, respectively. The comparison unit 60 compares the values calculated by the absolute value sum calculation units 51 to 54, and determines the prediction mode having the smallest prediction value as the optimum prediction mode.

  In addition, there exists a thing like patent document 1 as another technique of the prediction mode determination part 11. FIG.

JP 2008-172581 A

Joint Video Team (JVT) of ISO / IEC MPEG and ITU-T VCEG "Text Description of Joint Model Reference Encoding Methods and Decoding Concealment Methods"

  The prediction mode determination unit 11 in FIG. 27 calculates the prediction error of each prediction mode, and performs two-dimensional Amadal transformation on the calculated value. An arithmetic circuit that performs the calculation of the two-dimensional Amadar transform performs a matrix calculation, and thus the circuit scale becomes very large. Therefore, the image coding apparatus provided with the prediction mode determination unit 11 of FIG. 27 has a large hardware scale, which increases the chip cost.

  According to an aspect of the present invention, moving image data is input, and at least one prediction mode among a plurality of prediction modes is used to predict moving image data in units of macroblocks from reference pixels around the macroblock, An image encoding device that encodes the input moving image data using the predicted moving image data, wherein a first prediction mode out of the plurality of prediction modes causes a first prediction from the reference pixel. A first prediction pixel generation unit that generates a prediction pixel; a pixel value of an input pixel; and a difference calculation unit that generates a first prediction error by obtaining a difference between the pixel value of the first prediction pixel; From the reference pixel, an orthogonal transformation value generation unit that orthogonally transforms the first prediction error value to generate a first prediction error conversion value and at least one second prediction mode of the plurality of prediction modes. A correction value calculation unit that calculates a correction value for a pixel value obtained by orthogonally transforming two prediction pixels and a pixel value obtained by orthogonal transformation of the first prediction pixel; the correction value; and the first prediction error conversion value In response, the conversion value generation unit that generates the second prediction error conversion value for the second prediction mode is compared with the values corresponding to the first and second prediction error conversion values, And a determination unit that determines a prediction mode.

  According to another aspect of the present invention, moving image data is input, and at least one prediction mode among a plurality of prediction modes is used to predict moving image data in units of macroblocks from reference pixels around the macroblock. An image encoding method for encoding the input moving image data using the predicted moving image data, wherein the first prediction mode is a first prediction mode out of the plurality of prediction modes. A first prediction error is generated by calculating a difference between a pixel value of the input pixel and a pixel value of the first prediction pixel, and orthogonally transforming the first prediction error value, A first prediction error conversion value is generated, and a pixel value obtained by orthogonally transforming the second prediction pixel from the reference pixel in at least one second prediction mode among the plurality of prediction modes, and the first prediction Pixel A correction value for the orthogonally transformed pixel value is calculated, and a second prediction error conversion value for the second prediction mode is generated according to the correction value and the first prediction error conversion value. In this image encoding method, the one prediction mode is determined by comparing values corresponding to the first and second prediction error conversion values.

  The image coding apparatus according to the present invention, when determining at least one prediction mode among a plurality of prediction modes, the first prediction pixel calculated from the first prediction pixel generated by the first prediction pixel generation unit and the input pixel. It is only necessary to provide at least one orthogonal transformation value generation unit that orthogonally transforms the prediction error value. For this reason, it is not necessary to provide the orthogonal transformation value production | generation part with a large circuit scale for several prediction modes.

  The image coding apparatus according to the present invention can prevent an increase in hardware scale.

3 is a prediction mode determination unit of the image encoding device according to the first embodiment; It is a figure for demonstrating a macroblock and its peripheral pixel (color difference component). It is a schematic diagram for demonstrating DC prediction. This is a matrix of DC predicted pixels (divided blocks 0 to 3). It is a matrix coefficient of Hadamard transform. This is a matrix coefficient after Hadamard transform of a matrix of DC prediction pixels (divided blocks 0 to 3). It is a schematic diagram for demonstrating vertical prediction. This is a matrix of vertical prediction pixels (divided blocks 0, 2). It is a matrix coefficient after Hadamard transform of a matrix of vertical prediction pixels (division blocks 0 and 2). This is a matrix coefficient of the vertical prediction correction value generated by the vertical prediction correction value calculation unit (divided blocks 0 and 2). It is a schematic diagram for demonstrating horizontal prediction. This is a matrix of horizontal prediction pixels (divided blocks 0, 1). This is a matrix coefficient after Hadamard transform of a matrix of horizontal prediction pixels (divided blocks 0, 1). This is a matrix coefficient of the horizontal prediction correction value generated by the horizontal prediction correction value calculation unit (divided blocks 0 and 1). It is a schematic diagram for demonstrating plane prediction. This is a matrix of plane prediction pixels (divided blocks 0, 1). It is a matrix of plane prediction pixels (divided blocks 2, 3). It is a matrix coefficient after Hadamard transform of a matrix of plane prediction pixels (divided blocks 0, 1). It is a matrix coefficient after Hadamard transform of a matrix of plane prediction pixels (division blocks 2, 3). It is a matrix coefficient of the plane prediction correction value generated by the plane prediction correction value calculation unit (division block 0). It is a figure for demonstrating a macroblock (16 pixels x 16 pixels) and its peripheral pixel (luminance component). It is a figure for demonstrating a macroblock (16 pixels x 16 pixels) and its peripheral pixel (each division | segmentation block: 8 pixels x 8 pixels). 10 is a prediction mode determination unit of the image encoding device according to the second embodiment. It is the prediction mode determination part of the conventional image coding apparatus. It is the prediction mode determination part of the conventional image coding apparatus. It is a block configuration of an image encoding device. It is the prediction mode determination part of the conventional image coding apparatus.

  Embodiment 1 of the Invention

  Hereinafter, a specific first embodiment to which the present invention is applied will be described in detail with reference to the drawings. In the first embodiment, the present invention is applied to a prediction mode determination unit of an image encoding device. The overall configuration of the image encoding apparatus according to the first embodiment is the same as that of the image encoding apparatus 1 shown in FIG. Moreover, in this Embodiment 1, the optimal intra prediction mode which a prediction mode determination part determines shall be determined from four types, DC prediction, vertical prediction, horizontal prediction, and plane prediction.

  FIG. 1 shows a configuration of a prediction mode determination unit 100 of the image encoding device according to the first embodiment. As shown in FIG. 1, the prediction mode determination unit 100 includes a DC prediction pixel generation unit 111, a vertical prediction correction value calculation unit 112, a horizontal prediction correction value calculation unit 113, a plane prediction correction value calculation unit 114, and a difference calculation. Unit 120, two-dimensional Hadamard transform value generation unit 130, correction value addition units 141 to 143, absolute value sum calculation units 151 to 154, and comparison unit 160. Note that the following description is based on the assumption that the prediction mode determination unit 100 performs intra prediction for color difference components.

  The DC prediction pixel generation unit 111 performs prediction pixel data (hereinafter referred to as “prediction pixel data”) predicted in the DC prediction mode from adjacent reference pixels corresponding to a divided block having a basic unit of 4 pixels × 4 pixels in a macroblock described later for intra prediction. , Referred to as a predicted pixel value).

  The difference calculation unit 120 calculates a difference (hereinafter referred to as a prediction error) between the predicted pixel value predicted in the DC prediction mode of the divided block and the pixel value of the input pixel.

  The two-dimensional Hadamard transform value generation unit 130 performs a two-dimensional Hadamard transform (orthogonal transform) process on the DC prediction error calculated by the difference calculation unit 120, and the converted value (hereinafter referred to as a DC prediction error conversion value or DC prediction as necessary). Output error conversion coefficient).

  The vertical prediction correction value calculation unit 112 calculates a correction value (hereinafter referred to as a vertical prediction correction value) for Hadamard transform of prediction errors between the vertical prediction and the DC prediction from the pixel value of the adjacent reference pixel and the prediction pixel value predicted in the DC prediction mode. Is equivalently calculated using a predetermined calculation formula.

  The horizontal prediction correction value calculation unit 113 calculates a correction value (hereinafter referred to as a horizontal prediction correction value) for the Hadamard transform of the prediction error between the horizontal prediction and the DC prediction from the pixel value of the adjacent reference pixel and the prediction pixel value predicted in the DC prediction mode. Is equivalently calculated using a predetermined calculation formula.

  The plane prediction correction value calculation unit 114 calculates a correction value (hereinafter referred to as plane prediction correction value) for Hadamard transform of prediction errors between plane prediction and DC prediction from the pixel value of the adjacent reference pixel and the prediction pixel value predicted in the DC prediction mode. Is equivalently calculated using a predetermined calculation formula. Details of operations of the vertical prediction correction value calculation unit 112, the horizontal prediction correction value calculation unit 113, and the plane prediction correction value calculation unit 114 will be described later.

  The correction value adding units 141 to 143 (correction value generating units) add the vertical prediction correction value, the horizontal prediction correction value, and the plane prediction correction value to the DC prediction error conversion coefficient, respectively, and obtain the conversion coefficient of each prediction mode. calculate.

  The absolute value sum calculation units 151 to 154 calculate the sum of absolute values of the DC prediction error conversion coefficients from the two-dimensional Hadamard transform value generation unit 130 and the conversion coefficients calculated by the correction value addition units 141 to 143, respectively. Is output as the evaluation value for.

  The comparison unit 160 determines the one with the smallest value from the evaluation values of the respective prediction modes output by the absolute value sum calculation units 151 to 154 as the optimum prediction mode.

  Next, the operation and operation theory of the prediction mode determination unit 100 will be described using mathematical formulas and drawings. First, a macroblock used in each prediction mode of general intra prediction will be described with reference to FIG. As shown in FIG. 2, the macroblock has divided blocks block0 to block3 each consisting of 4 pixels × 4 pixels. Adjacent reference pixels indicated by hatching are adjacent to the periphery of such a macroblock. As the adjacent reference pixels, there are adjacent pixels p0 to p7 in the horizontal direction of the macroblock, and adjacent pixels q0 to q7 in the vertical direction. In addition, there is an adjacent pixel pq at the upper left of the macroblock, that is, at a location in contact with the adjacent pixels p0 and q0. The values pq, p0 to p7, and q0 to q7 shown for each pixel indicate the pixel name and the pixel value of that pixel. Further, each pixel arranged in a matrix of macroblocks is assumed to be M (x, y) (x = 0 to 7, y = 0 to 7).

  The DC prediction pixel generation unit 111 calculates a prediction pixel value of each divided block from the adjacent reference pixels. Here, FIG. 3 shows a conceptual diagram of intra prediction in the DC prediction mode. As shown in the conceptual diagram of FIG. 3, the DC prediction pixel generation unit 111 calculates a prediction pixel value of each divided block based on an average value of adjacent reference pixel values corresponding to each divided block as DC prediction. The predicted pixel values P (x, y) of the DC predicted values DC (k) (k = 0 to 3) corresponding to the divided blocks block 0 to block 3 are as follows.

In the case of the DC prediction value DC (0) of the divided block,
P (x, y) = (p0 + p1 + p2 + p3 + q0 + q1 + q2 + q3 + 4) / 8 (x = 0-3, y = 0-3)
It becomes.

In the case of the DC prediction value DC (1) of the divided block,
P (x, y) = (p4 + p5 + p6 + p7 + 2) / 4 (x = 4-7, y = 0-3)
It becomes.

In the case of the DC prediction value DC (2) of the divided block,
P (x, y) = (q4 + q5 + q6 + q7 + 2) / 4 (x = 0-3, y = 4-7)
It becomes.

In the case of the DC prediction value DC (3) of the divided block,
P (x, y) = (p4 + p5 + p6 + p7 + q4 + q5 + q6 + q7 + 4) / 8 (x = 4-7, y = 4-7)
It becomes. FIG. 4 shows a prediction pixel matrix P_DC (k) of prediction values obtained by DC prediction of each divided block as described above.

  The difference calculation unit 120 calculates the difference between the prediction pixel matrix P_DC (k) calculated by the DC prediction pixel generation unit 111 and the pixel value of the input pixel. Hereinafter, unless otherwise specified, the prediction pixel matrix P_DC (k) is referred to as “prediction pixel P_DC”.

Here, as described above, the prediction mode determination unit 100 according to the first embodiment has the least difference between the input pixel value and the prediction pixel value obtained by DC prediction, vertical prediction, horizontal prediction, and plane prediction. Is determined as the optimum mode. In addition, the prediction mode determination unit 100 uses a two-dimensional Hadamard transform for this determination. When the Hadamard transformation matrix is H and the prediction difference is D, the prediction difference matrix TD after Hadamard transformation is
TD = H × D × H
It becomes.

  The Hadamard transform matrix H is a matrix having coefficients as shown in FIG. Here, FIG. 6 shows the frequency axis matrix T_DC (k) after the two-dimensional Hadamard transform of the prediction pixel P_DC of FIG. As shown in FIG. 6, the matrix T_DC (k) after the two-dimensional Hadamard transform is substantially one coefficient 16DC (k). Hereinafter, the matrix T_DC (k) (k = 0 to 3) after the two-dimensional Hadamard transform of the predicted pixel matrix P_DC (k) is referred to as “predicted pixel Hadamard transform T_DC”.

Here, a matrix of input pixel values for which the difference calculation unit 120 obtains a difference from the predicted pixel P_DC is referred to as “O” (hereinafter simply referred to as the input pixel O). Then, assuming that the Hadamard transform result of the input pixel O is T_Orig, the Hadamard transform TD_DC of the prediction error of the DC prediction can be expressed by the following equation (1).
TD_DC
= H * (OP-DC) * H = H * O * H-H * P_DC * H
= T_Orig-T_DC (1)

  Next, in order to explain the operation theory of the vertical prediction correction value calculation unit 112, the horizontal prediction correction value calculation unit 113, and the plane prediction correction value calculation unit 114, intra prediction in each prediction mode of vertical prediction, horizontal prediction, and plane prediction is performed. explain.

First, FIG. 7 shows a conceptual diagram of intra prediction in the vertical prediction mode. As shown in the conceptual diagram of FIG. 7, in the vertical prediction mode, the predicted pixel value of each pixel is determined based on the adjacent pixel values p0 to p7 in the horizontal direction. Therefore, the predicted pixel value P (x, y) by vertical prediction is
P (0, y) = p0 (y = 0-7)
P (1, y) = p1 (y = 0-7)
P (2, y) = p2 (y = 0-7)
P (3, y) = p3 (y = 0-7)
P (4, y) = p4 (y = 0-7)
P (5, y) = p5 (y = 0-7)
P (6, y) = p6 (y = 0-7)
P (7, y) = p7 (y = 0-7)
It becomes.

  Here, the divided block block 0 will be described by paying attention to the divided block block 0 on behalf of the divided blocks block 0 to block 3. FIG. 8 shows a prediction pixel matrix P_VERT (0) of prediction values obtained by vertically predicting the divided block block0. Note that since the divided block block2 has the same value as the divided block block0, the prediction pixel matrix P_VERT (2) also has the same value as P_VERT (0). Further, the prediction pixel matrices P_VERT (1) and P_VERT (3) of the divided blocks block1 and block3 are obtained by replacing p0 to p3 of the prediction pixel matrix P_VERT (0) with values of p4 to p7, respectively.

Here, the prediction pixel matrix of FIG. 8 is two-dimensional Hadamard transformed. FIG. 9 shows a frequency axis matrix T_VERT (0) after the two-dimensional Hadamard transform of the prediction pixel matrix of FIG. However,
A1 = (p0 + p1 + p2 + p3) × 4
A2 = (p0 + p1-p2-p3) × 4
A3 = (p0−p1−p2 + p3) × 4
A4 = (p0−p1 + p2−p3) × 4
And

As shown in FIG. 9, the matrix T_VERT (0) after the two-dimensional Hadamard transform is “0” except for the four low-frequency coefficients. The divided block block2 has the same value. In addition, matrices T_VERT (1) and T_VERT (3) after the two-dimensional Hadamard transform of block1 and block3 are
A1 = (p4 + p5 + p6 + p7) × 4
A2 = (p4 + p5-p6-p7) × 4
A3 = (p4-p5-p6 + p7) × 4
A4 = (p4-p5 + p6-p7) × 4
Other than that, it is the same as P_VERT (0).

  In the following, unless otherwise specified, the prediction pixel matrix P_VERT (k) (k = 0 to 3) is “vertical prediction pixel P_VERT”, and the matrix after the two-dimensional Hadamard transform of the prediction pixel matrix P_VERT (k) T_VERT (k) (k = 0 to 3) is referred to as “predicted pixel Hadamard transform T_VERT”.

Here, the Hadamard transform TD_VERT of the prediction error of the vertical prediction can be expressed by the following equation (2).
TD_VERT
= H * (OP-VERT) * H = H * O * H-H * P_VERT * H
= T_Orig-T_VERT (2)

Here, the difference between the Hadamard transform TD_VERT of the prediction error of the vertical prediction and the Hadamard transform TD_DC of the prediction error of the DC prediction is obtained. This difference can be expressed by the following formula (3) by formula (1) and formula (2).
TD_VERT-TD_DC
= (T_Orig-T_VERT)-(T_Orig-T_DC)
= T_DC-T_VERT (3)

When the Hadamard transform TD_VERT of the prediction error of the vertical prediction is obtained based on this equation (3), the following equation (4) is obtained.
TD_VERT
= TD_DC + T_DC-T_VERT
= TD_DC + ΔTD_VERT (4)
Here, ΔTD_VERT in Expression (4) is set as the vertical prediction correction value.

  The vertical prediction correction value calculation unit 112 calculates prediction pixel Hadamard transforms T_VERT and T_DC from the vertical prediction pixel P_VERT generated from the adjacent reference pixel and the DC prediction pixel P_DC generated by the DC prediction pixel generation unit 111. Then, a vertical prediction correction value ΔTD_VERT in the above equation (4) is calculated from the predicted pixel Hadamard transform T_VERT and T_DC.

  Here, as described above, the prediction pixel Hadamard transform T_DC except one coefficient of the low frequency is "0", the prediction pixel Hadamard transform T_VERT is other than four coefficients of low frequency "0". Therefore, the calculation of the vertical prediction correction value ΔTD_VERT is “0” except for the four low-frequency coefficients among the 16 coefficients included in the determinant, and it is sufficient to pay attention to only those four coefficients. Therefore, the amount of calculation can be reduced to about ¼ compared with the case where all 16 coefficients are performed. For example, the vertical prediction correction value ΔTD_VERT (0) of the divided block block0 is derived by calculating the matrix difference as shown in FIG.

Next, FIG. 11 shows a conceptual diagram of intra prediction in the horizontal prediction mode. As shown in the conceptual diagram of FIG. 11, in the horizontal prediction mode, the predicted pixel value of each pixel is determined by the adjacent pixel values q0 to q7 in the vertical direction. Therefore, the predicted pixel value P (x, y) by horizontal prediction is
P (x, 0) = q0 (x = 0 to 7)
P (x, 1) = q1 (x = 0 to 7)
P (x, 2) = q2 (x = 0 to 7)
P (x, 3) = q3 (x = 0 to 7)
P (x, 4) = q4 (x = 0 to 7)
P (x, 5) = q5 (x = 0 to 7)
P (x, 6) = q6 (x = 0 to 7)
P (x, 7) = q7 (x = 0-7)
It becomes.

  Here, the divided block block 0 will be described by paying attention to the divided block block 0 on behalf of the divided blocks block 0 to block 3. FIG. 12 shows a prediction pixel matrix P_HORI (0) of prediction values obtained by vertically predicting the divided block block0. Since the divided block block1 has the same value as the divided block block0, the prediction pixel matrix P_HORI (1) also has the same value as P_HORI (0). Further, the prediction pixel matrices P_HORI (2) and P_VERT (3) of the divided blocks block2 and block3 are obtained by replacing q0 to q3 of the prediction pixel matrix P_HORI (0) with the values of q4 to q7, respectively.

Here, the prediction pixel matrix of FIG. 12 is two-dimensional Hadamard transformed. FIG. 13 shows a frequency axis matrix T_HORI (0) after the two-dimensional Hadamard transform of the prediction pixel matrix of FIG. However, B1 = (q0 + q1 + q2 + q3) × 4
B2 = (q0 + q1-q2-q3) × 4
B3 = (q0−q1−q2 + q3) × 4
B4 = (q0−q1 + q2−q3) × 4
And As shown in FIG. 13, the matrix T_HORI (0) after the two-dimensional Hadamard transform is “0” except for the four low-frequency coefficients. The block block1 has the same value.

In addition, the matrices T_HORI (2) and T_HORI (3) after the two-dimensional Hadamard transform of block2 and block3 are
B1 = (q4 + q5 + q6 + q7) × 4
B2 = (q4 + q5-q6-q7) × 4
B3 = (q4-q5-q6 + q7) × 4
B4 = (q4-q5 + q6-q7) × 4
Is the same as P_HORI (0) except that In the following, unless otherwise specified, the prediction pixel matrix P_HORI (k) (k = 0 to 3) of horizontal prediction is “horizontal prediction pixel P_HORI”, and the two-dimensional Hadamard transform of the prediction pixel matrix P_HORI (k) The subsequent matrix T_HORI (k) (k = 0 to 3) is referred to as “predicted pixel Hadamard transform T_HORI”.

Here, the Hadamard transform TD_HORI of the prediction error of horizontal prediction can be expressed by the following equation (5).
TD_HORI
= H * (OP-HORI) * H = H * O * H-H * P_HORI * H
= T_Orig-T_HORI (5)

Here, the difference between the Hadamard transform TD_HORI of the prediction error of horizontal prediction and the Hadamard transform TD_DC of the prediction error of DC prediction is obtained. This difference can be expressed by the following equation (6) using equations (1) and (5).
TD_HORI-TD_DC
= (T_Orig-T_HORI)-(T_Orig-T_DC)
= T_DC-T_HORI (6)

When the Hadamard transform TD_HORI of the prediction error of horizontal prediction is obtained based on this equation (6), the following equation (7) is obtained.
TD_HORI
= TD_DC + T_DC-T_HORI
= TD_DC + ΔTD_HORI (7)
Here, ΔTD_HORI in Expression (7) is set as the horizontal prediction correction value.

  The horizontal prediction correction value calculation unit 113 calculates prediction pixel Hadamard transforms T_HORI and T_DC from the horizontal prediction pixel P_HORI generated from the adjacent reference pixels and the DC prediction pixel P_DC generated by the DC prediction pixel generation unit 111. Then, a horizontal prediction correction value ΔTD_HORI in the above equation (7) is calculated from the predicted pixel Hadamard transform T_HORI and T_DC.

  Here, as described above, the predicted pixel Hadamard transform T_DC is “0” except for one coefficient in the low band, and the predicted pixel Hadamard transform T_HORI is “0” except for the four coefficients in the low band. Therefore, the calculation of the horizontal prediction correction value ΔTD_HORI is “0” except for the four coefficients in the low band among the 16 coefficients included in the determinant, and it is only necessary to pay attention to only those four coefficients. Therefore, the amount of calculation can be reduced to ¼ compared to the case where all 16 coefficients are performed. For example, the horizontal prediction correction value ΔTD_HORI (0) of the divided block block0 is derived by calculating the matrix difference as shown in FIG.

Next, FIG. 15 shows a conceptual diagram of intra prediction in the planar prediction mode. In the planar prediction mode, the predicted value predicted value P (x, y) (x = 0 to 7, y = 0 to 7) of each pixel is
P (x, y) = (a + b × (x−3) + c × (y−3) +16) / 32
It becomes. However,
a = (p7 + q7) × 16
b = (34 × H + 32) / 64
c = (34 × V + 32) / 64
H = (p7-pq) × 4 + (p6-p0) × 3 + (p5-p1) × 2 + (p4-p2)
V = (q7-pq) × 4 + (q6-q0) × 3 + (q5-q1) × 2 + (q4-q2)
It is. Here, FIGS. 16A and 16B show prediction pixel matrices P_PLAN (0) and P_PLAN (1) of prediction values obtained by plane prediction of the divided blocks block0 and block1, respectively. FIGS. 17A and 17B show prediction pixel matrices P_PLAN (2) and P_PLAN (3) of predicted values obtained by plane prediction of the divided blocks block2 and block3.

  Here, the prediction pixel matrix of FIG. 16A, FIG. 16B, FIG. 17A, and FIG. FIGS. 18A and 18B show frequency axis matrices T_PLAN (0) and T_PLAN (1) after the two-dimensional Hadamard transform of the prediction pixel matrix of FIGS. 16A and 16B. FIGS. 19A and 19B show frequency axis matrices T_PLAN (2) and T_PLAN (3) after the two-dimensional Hadamard transform of the prediction pixel matrix of FIGS. 17A and 17B.

  As shown in FIG. 18A, FIG. 18B, FIG. 19A, and FIG. 19B, the matrix T_PLAN (k) (k = 0 to 3) after the two-dimensional Hadamard transform has a low frequency range. Other than these five coefficients are “0”. In the following description, unless otherwise specified, the prediction pixel matrix P_PLAN (k) (k = 0 to 3) for plane prediction is referred to as “plane prediction pixel P_PLAN”, and the two-dimensional Hadamard transform of the prediction pixel matrix P_PLAN (k). The subsequent matrix T_PLAN (k) (k = 0 to 3) is referred to as “predicted pixel Hadamard transform T_PLAN”.

Here, Hadamard transform TD_PLAN of prediction error of plane prediction can be expressed by the following equation (8).
TD_PLAN
= H * (OP-PLAN) * H = H * O * H-H * P_PLAN * H
= T_Orig-T_PLAN (8)

Here, the difference between the Hadamard transform TD_PLAN of the prediction error of plane prediction and the Hadamard transform TD_DC of the prediction error of DC prediction is obtained. This difference can be expressed by the following formula (9) by formula (1) and formula (8).
TD_PLAN-TD_DC
= (T_Orig-T_PLAN)-(T_Orig-T_DC)
= T_DC-T_PLAN (9)

When the Hadamard transform TD_PLAN of the prediction error of plane prediction is obtained based on this equation (9), the following equation (10) is obtained.
TD_PLAN
= TD_DC + T_DC-T_PLAN
= TD_DC + ΔTD_PLAN (10)
Here, ΔTD_PLAN in Expression (10) is set as a plane prediction correction value.

  The plane prediction correction value calculation unit 114 calculates prediction pixel Hadamard transforms T_PLAN and T_DC from the plane prediction pixel P_PLAN generated from the adjacent reference pixels and the DC prediction pixel P_DC generated by the DC prediction pixel generation unit 111. Then, the plane prediction correction value ΔTD_PLAN in the above equation (10) is calculated from the predicted pixel Hadamard transform T_PLAN and T_DC.

  Here, as described above, the predicted pixel Hadamard transform T_DC is “0” except for one coefficient in the low band, and the predicted pixel Hadamard transform T_PLAN is “0” except for the five coefficients in the low band. Therefore, the calculation of the plane prediction correction value ΔTD_PLAN is “0” except for the five low-frequency coefficients among the 16 coefficients included in the determinant, and it is sufficient to pay attention to only those five coefficients. Therefore, the amount of calculation can be reduced to 5/16 compared with the case where all 16 coefficients are performed. For example, the plane prediction correction value ΔTD_PLAN (0) of the divided block block0 is derived by calculating the matrix difference as shown in FIG.

  The correction value addition unit 141 calculates the sum of the vertical prediction correction value ΔTD_VERT output from the vertical prediction correction value calculation unit 112 and the Hadamard transform TD_DC of the prediction error of DC prediction output from the two-dimensional Hadamard transform value generation unit 130. calculate. Since this calculation corresponds to Equation (4), the correction value adding unit 141 outputs the Hadamard transform TD_VERT of the prediction error of the vertical prediction as a result.

  The correction value addition unit 142 calculates the sum of the horizontal prediction correction value ΔTD_HORI output from the horizontal prediction correction value calculation unit 113 and the Hadamard transform TD_DC of the prediction error of DC prediction output from the two-dimensional Hadamard transform value generation unit 130. calculate. Since this calculation corresponds to Equation (7), the correction value adding unit 142 outputs the Hadamard transform TD_HORI of the prediction error of the horizontal prediction as a result.

  The correction value addition unit 143 calculates the sum of the plane prediction correction value ΔTD_PLAN output from the plane prediction correction value calculation unit 114 and the Hadamard transform TD_DC of the prediction error of the DC prediction output from the two-dimensional Hadamard transform value generation unit 130. calculate. Since this calculation corresponds to Expression (10), the Hadamard transform TD_PLAN of the prediction error of plane prediction is output from the correction value adding unit 143 as a result.

  Absolute value sum calculating section 151 to 154, calculates Hadamard transform TD_DC of prediction errors respectively above, TD_VERT, TD_HORI, the absolute value sum of TD_PLAN. And the comparison part 160 compares the absolute value sum of each prediction mode which the absolute value sum calculation parts 151-154 output, and determines with the prediction mode of the minimum value being the optimal prediction mode.

  Here, the prediction mode determination unit 11 of the conventional image encoding device in FIG. 27 calculates the prediction error of each prediction mode, and the two-dimensional Amadal transform value generation units 41 to 44 perform 2 for the calculated values. Dimensional Amadar transformation was performed. In this case, the calculation of the two-dimensional Amadal transform must be executed for all 16 coefficients of the prediction error value of each prediction mode. For this reason, the calculation amount has become very large. In addition, each of the arithmetic circuits of the two-dimensional Amadal transformation value generation units 41 to 44 that perform the calculation of the two-dimensional Amadal transformation performs a matrix calculation, so that the circuit scale is large, and the hardware scale of the prediction mode determination unit 11 is large. There was a problem of becoming very large.

  However, the prediction mode determination unit 100 of the image coding apparatus of the first embodiment calculates only the prediction error of the DC prediction mode, two-dimensional Hadamard transform value generating unit 130 with respect to the calculated value is a two-dimensional Hadamard Just do the conversion. Therefore, it is not necessary to perform the two-dimensional Amadal transformation of the prediction error for the prediction modes of vertical prediction, horizontal prediction, and planar prediction, and the number of arithmetic circuits can be reduced accordingly. When the vertical prediction correction value calculation unit 112, the horizontal prediction correction value calculation unit 113, and the plane prediction correction value calculation unit 114 calculate the vertical prediction correction value ΔTD_VERT, the horizontal prediction correction value ΔTD_HORI, and the plane prediction correction value ΔTD_PLAN, Although two-dimensional Amadal transformation is performed, the amount of calculation is small and the scale of the arithmetic circuit can be small because the coefficient to be calculated may be small as described above. For this reason, the hardware scale of the prediction mode determination part 100 can be reduced conventionally.

  The above description is based on the standard H.264 standard. Although the prediction mode determination of the H.264 color difference component has been described, the same method as described above can be applied to the luminance component. H. According to the H.264 standard, luminance components include macroblocks of 16 pixels × 16 pixels and 4 pixels × 4 pixels as shown in FIG. The system of the first embodiment can also be applied to macroblocks of 16 pixels × 16 pixels or 4 pixels × 4 pixels. Also, as shown in FIG. 22, the method of the first embodiment can be applied even when one block is expanded to 8 pixels × 8 pixels.

  Embodiment 2 of the Invention

  Hereinafter, a specific second embodiment to which the present invention is applied will be described in detail with reference to the drawings. In the second embodiment, as in the first embodiment, the present invention is applied to a prediction mode determination unit of an image encoding device. As in the first embodiment, the overall configuration of the image coding apparatus according to the second embodiment is the same as that of the image coding apparatus 1 shown in FIG. In the second embodiment, the optimum intra prediction mode determined by the prediction mode determination unit is determined from four types of DC prediction, vertical prediction, horizontal prediction, and plane prediction, and other prediction modes other than that. And

  FIG. 23 illustrates a configuration of the prediction mode determination unit 200 of the image encoding device according to the second embodiment. As shown in FIG. 23, the prediction mode determination unit 200 includes a DC prediction pixel generation unit 111, a vertical prediction correction value calculation unit 112, a horizontal prediction correction value calculation unit 113, a plane prediction correction value calculation unit 114, and a difference calculation. Unit 120, two-dimensional Hadamard transform value generation unit 130, correction value addition units 141 to 143, absolute value sum calculation units 151 to 154, comparison unit 160, other prediction pixel generation unit 210, selector 220, And a storage unit 230.

  In addition, the structure which attached | subjected the code | symbol same as FIG. 1 among the code | symbols shown in FIG. 23 has shown the structure same as or similar to FIG. The difference between the second embodiment and the first embodiment is that the prediction mode determination unit 200 is optimal from prediction modes including other prediction modes other than the four types of DC prediction, vertical prediction, horizontal prediction, and plane prediction. The prediction mode is determined. For this reason, the second embodiment will focus on the differences from the first embodiment, and the description of the same parts as the first embodiment will be omitted.

  The other prediction pixel generation unit 210 generates prediction pixel data (hereinafter referred to as a prediction pixel value) from the reference pixel for prediction in other prediction modes other than DC prediction, vertical prediction, horizontal prediction, and plane prediction. Other prediction modes include modes such as modes 3 to 8 in inter prediction and intra prediction of luminance components.

  The selector 220 selects either the output from the DC prediction pixel generation unit 111 or the output from the other prediction pixel generation unit 210 and outputs it to the difference calculation unit 120.

  The storage unit 230 stores the absolute value sum value of the Hadamard transform of the prediction error or the absolute value sum of the Hadamard transform of the DC prediction error, which is output by the absolute value sum calculation unit 151, The stored value is output to the comparison unit 160.

  Note that the selector 220 selects all inputs in a time division manner in order to perform comparative evaluation of the plurality of prediction modes that are the inputs. The storage unit 230 also stores the value output from the absolute value sum calculation unit 151 according to the input selected by the selector 220.

  For example, first, if the output from the other predicting pixel generator 210 selector 220 selects absolute value sum calculating unit 151 other prediction mode storage unit 230 the absolute value sum of Hadamard transform of the prediction error by calculating the the Remember once. After that, when the selector 220 selects the output from the DC prediction pixel generation unit 111, the storage unit 230 does not store the absolute value sum value of the Hadamard transform of the DC prediction error calculated by the absolute value sum calculation unit 151. The calculation result from the absolute value sum calculation unit 151 is directly input to the comparison unit 160 and compared.

  And vice versa, i.e., first case where the selector 220 the output from the selector 220 is DC prediction pixel generation unit 111 selects, once the value of the sum of absolute values of Hadamard transform DC prediction error absolute value sum calculation unit 151 calculates The storage unit 230 stores it. Thereafter, the output from the other predicting pixel generator 210 when the selector 220 selects, not the absolute value sum calculating unit 151 other storage unit 230 the absolute value sum of Hadamard transform of the prediction error due to the prediction mode to be calculated is stored Instead, the calculation result from the absolute value sum calculation unit 151 may be directly input to the comparison unit 160 for comparison.

  When the comparison unit 160 has a function of temporarily storing the absolute value sum of the Hadamard transform of the prediction error or the absolute value sum of the Hadamard transform of the DC prediction error in the other prediction modes, the storage unit 230 may be deleted.

  The operation of the prediction mode determination unit 200 having such a configuration will be described. In addition, the memory | storage part 230 assumes the case where the absolute value sum of Hadamard transformation of the prediction error by another prediction mode is memorize | stored.

  First, the other prediction pixel production | generation part 210 produces | generates a prediction pixel value from an adjacent reference pixel. The selector 220 outputs the data output from the other prediction pixel generation unit 210 to the difference calculation unit 120. The difference calculation unit 120 calculates the difference between the pixel value of the input pixel and the prediction pixel value output from the other prediction pixel generation unit 210 and outputs the difference as prediction error data. The two-dimensional Hadamard transform value generation unit 130 performs a two-dimensional Hadamard transform process on the prediction error and outputs it to the absolute value sum calculation unit 151. The absolute value sum calculation unit 151 calculates the absolute value sum of the output from the two-dimensional Hadamard transform value generation unit 130. The storage unit 230 stores the value calculated by the absolute value sum calculation unit 151.

  Thereafter, performed the operation described in the first embodiment, each DC prediction from the absolute value sum calculating section 151 to 154, vertical prediction, the sum of absolute values of horizontal prediction, Hadamard transform of the prediction error due to the prediction mode of the plane prediction Calculated. The comparison unit 160 calculates the sum of absolute values of Hadamard transform of prediction errors in the prediction modes of DC prediction, vertical prediction, horizontal prediction, and plane prediction, and Hadamard transform of prediction errors in other prediction modes stored in the storage unit 230. Is compared with the sum of absolute values of. And it determines with the prediction mode with the smallest value being the optimal prediction mode among the absolute value sum of the Hadamard transform of the prediction error of these five types of prediction modes.

  Here, in the technique disclosed in Patent Document 1, the configuration of the prediction mode determination unit 300 in the case where other prediction modes other than the four types of DC prediction, vertical prediction, horizontal prediction, and plane prediction are added is shown in FIG. Show. Prediction mode determination unit 300 shown in FIG. 24, a two-dimensional Hadamard transform value generating unit 310, a DC prediction Hadamard transform value generating unit 311, a vertical prediction Hadamard transform value generating unit 312, a horizontal prediction Hadamard transform value generating unit 313 , a plane prediction Hadamard transform value generating unit 314, and the other predicted Hadamard transform value generating unit 315, and other prediction pixel generation unit 320, a difference calculation unit 331 to 335, an absolute value sum calculating section 341 to 345, compared Part 350.

  The operation of the prediction mode determination unit 300 will be briefly described. First, the two-dimensional Hadamard transform value generation unit 310 calculates the two-dimensional Hadamard transform of the input pixel and generates the converted value. DC prediction Hadamard transform value generating unit 311, vertical prediction Hadamard transform value generating unit 312, horizontal prediction Hadamard transform value generating unit 313, plane prediction Hadamard transform value generating unit 314, a two-dimensional Hadamard transform for each predicted value from the adjacent reference pixels Calculate and generate the converted value.

  The other prediction pixel generation unit 320 generates a prediction pixel in a prediction mode other than the DC prediction, vertical prediction, horizontal prediction, and plane prediction. The other predicted Hadamard transform value generation unit 315 calculates the two-dimensional Hadamard transform of the predicted pixel generated by the other predicted pixel generation unit 320 and generates the converted value.

  The difference calculation units 331 to 335 calculate the difference between the Hadamard transform value generated by the Hadamard transform value generation units 311 to 315 of each prediction mode described above and the Hadamard transform value of the input pixel, and calculate the prediction error of each prediction mode. Output. The absolute value sum calculation units 341 to 345 calculate the absolute value sum of the prediction errors in these prediction modes. The comparison unit 350 compares the absolute value sum values of the prediction errors of the respective prediction modes calculated by the absolute value sum calculation units 341 to 345. And the comparison part 350 determines with the prediction mode with the smallest value being the optimal prediction mode among the absolute value sums of the Hadamard transform of the prediction error of these five types of prediction modes.

  The above is the operation of the prediction mode determination unit 300. Here, the added Hadamard transform DC prediction of other prediction modes, vertical prediction, horizontal prediction, if not readily simplified coefficients as planar prediction, and other prediction pixel generation unit 320 as described above generates The other predicted Hadamard transform value generation unit 315 is required for Hadamard transform of the predicted pixel.

  However, in the prediction mode determination unit 200 according to the second embodiment, since the two-dimensional Hadamard transform value generation unit 130 performs the difference between the prediction pixel generated by the other prediction pixel generation unit 210 and the input pixel, the prediction mode determination is performed. The components corresponding to the other predicted Hadamard transform value generation unit 315 in the unit 300 can be reduced.

  Furthermore, in the technique disclosed in Patent Document 1, input pixels used for calculation of prediction errors in DC prediction, vertical prediction, horizontal prediction, and plane prediction modes, and prediction errors in other prediction modes are calculated. When the Hadamard transform process with the input pixel to be used is performed at different timings, a configuration like the prediction mode determination unit 400 illustrated in FIG. 25 is necessary. The prediction mode determination unit 400, and the other of the prediction pixel generation unit 320 of the prediction mode determination unit 300, a difference calculation unit 335, instead of the absolute value sum calculation unit 345 is eliminated, a selector 411 and 412, storage unit 421, 422 Is added.

  The operation in this case is as follows. First, the selector 410 selects the output of the prediction pixel generated by the other prediction pixel generation unit 320 and outputs it to the two-dimensional Hadamard transform value generation unit 310. Then, the two-dimensional Hadamard transform value generation unit 310 calculates the two-dimensional Hadamard transform of the predicted pixel. The storage unit 420 stores Hadamard transform values that are the calculation results. Next, the two-dimensional Hadamard transform value generation unit 310 calculates the two-dimensional Hadamard transform of the input pixel selected by the selector 411. Difference calculation unit 335 calculates a Hadamard transform value of the prediction pixel of other prediction modes storage unit 420 stores the difference between the two-dimensional Hadamard transform value of the input pixel, and outputs the prediction error. The selector 412 selects the output from the difference calculation unit 335 and outputs it to the absolute value sum calculation unit 341. The storage unit 422 stores the absolute value sum value of prediction errors in other prediction modes calculated by the absolute value sum calculation unit 341.

  Thereafter, the absolute value sums of the prediction errors of DC prediction, vertical prediction, horizontal prediction, and plane prediction are output from the absolute value sum calculation units 341 to 344 by the operation described in the prediction mode determination unit 300, respectively. The comparison unit 350 compares the value output from the absolute value sum calculation units 341 to 344 with the value stored in the storage unit 422. And the comparison part 350 determines with the prediction mode with the smallest value being the optimal prediction mode among the absolute value sum of the Hadamard transform of the prediction error of these 5 types of prediction modes.

  In such a prediction mode determination unit 400, the Hadamard transform value generation unit 315 is reduced with respect to the prediction mode determination unit 300, but a storage unit 421 is added. Since the storage unit 421 stores matrix coefficients, the circuit scale increases. Therefore, the prediction mode determination unit 400 also has a larger circuit scale than the prediction mode determination unit 200 of the second embodiment.

  As described above, even when other prediction modes other than DC prediction, vertical prediction, horizontal prediction, and plane prediction are added, the prediction mode determination unit 200 increases the circuit scale with respect to the prediction mode determination units 300 and 400 described above. It is possible to suppress.

  Note that the present invention is not limited to the above-described embodiment, and can be changed as appropriate without departing from the spirit of the present invention. For example, in the first and second embodiments, a prediction pixel for DC prediction is generated, correction values for vertical prediction, horizontal prediction, and plane prediction are calculated based on the prediction pixel, and the correction value and a prediction error value for DC prediction are calculated. I was seeking the sum of

  Here, instead of DC prediction, a prediction pixel of another prediction mode is generated, a correction value of the remaining prediction mode is calculated based on the prediction pixel, and orthogonal correction of the correction value and the prediction error of the other prediction mode is performed. You may ask for the sum with a result. For example, a prediction pixel for vertical prediction may be generated, a correction value for DC prediction, horizontal prediction, and plane prediction may be calculated based on the prediction pixel, and a sum of the correction value and an orthogonal transformation result for vertical prediction may be obtained. That is, the reference prediction mode can be set arbitrarily. Further, the number of prediction modes serving as the reference is also arbitrary. For example, two prediction methods, DC prediction and vertical prediction, are used as reference prediction modes. However, in this case, the correction values of the remaining prediction modes that have not become the reference are calculated from the prediction pixels of the reference prediction mode.

  Furthermore, the number of prediction modes for calculating correction values is also arbitrary. For example, a prediction pixel of DC prediction is generated, correction values for two prediction modes of vertical prediction and horizontal prediction are calculated based on the prediction pixel, and the sum of the correction value and the orthogonal transformation result of the prediction error of DC prediction is calculated. You may make it ask. For planar prediction in which no correction value is calculated, a prediction pixel may be generated and a prediction error may be calculated and orthogonal transformation may be performed in the same manner as in DC prediction.

100, 200 Prediction mode determination unit 111 DC prediction pixel generation unit 112 Vertical prediction correction value calculation unit 113 Horizontal prediction correction value calculation unit 114 Plane prediction correction value calculation unit 120 Difference calculation unit 130 Two-dimensional Hadamard transform value generation units 141 to 143 Correction Value addition units 151 to 154 Absolute value sum calculation unit 160 Comparison unit 210 Other predicted pixel generation unit 220 Selector 230 Storage unit

Claims (1)

Moving picture data is input, using at least one prediction mode of a plurality of prediction modes, moving picture data in units of macroblocks is predicted from reference pixels around the macroblock, and the predicted moving picture data is And an image encoding device for encoding the input moving image data,
A first prediction pixel generation unit configured to generate a first prediction pixel from the reference pixel according to a first prediction mode of the plurality of prediction modes;
A difference calculation unit that generates a first prediction error by calculating a difference between a pixel value of an input pixel and a pixel value of the first prediction pixel;
An orthogonal transform value generation unit that orthogonally transforms the first prediction error value to generate a first prediction error transform value;
Correction value for a pixel value obtained by orthogonally transforming the second predicted pixel from the reference pixel and a pixel value obtained by orthogonally transforming the first predicted pixel in at least one second prediction mode of the plurality of prediction modes A correction value calculation unit for calculating
A conversion value generation unit that generates a second prediction error conversion value for the second prediction mode according to the correction value and the first prediction error conversion value;
An image coding apparatus comprising: a determination unit that compares the values according to the first and second prediction error conversion values to determine the one prediction mode.

Images