JP2009272727A - Transformation method based on directivity of prediction error, image-encoding method and image-decoding method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a moving image-encoding method for which the power concentration of a transformation coefficient is improved by orthogonal transformation, corresponding to the direction of in-screen prediction. <P>SOLUTION: The moving image-encoding method for encoding input image signals includes: a step of generating predicted error signals (116), indicating the difference value among prediction signals (121), generated according to a predetermined prediction mode and the input image signals (115); an orthogonal transformation step of orthogonally transforming the prediction error signals by a first transformation method, depending on the prediction direction of the prediction mode or a second transformation method which is independent of the prediction direction and generating the transformation coefficients; a quantization step of executing quantization processing on the transformation coefficients and generating quantized transformation coefficients; and an encoding processing step of executing entropy-encoding processing to the quantized transformation coefficient (117) and transformation information (118) indicating whether it is the first transformation method or the second transformation method and generating encoded data (124). <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a moving image encoding device, a moving image encoding method, a moving image decoding device, and a moving image decoding method, and in particular, improves power concentration of transform coefficients by orthogonal transform corresponding to the direction of intra prediction. The present invention relates to a moving image encoding device, a moving image encoding method, a moving image decoding device, and a moving image decoding method, which can obtain an effect of improving encoding efficiency.

H. In a moving picture coding system such as H.264 / MPEG-4 AVC (hereinafter referred to as H.264), a discrete cosine transform (hereinafter referred to as DCT) / inverse discrete cosine transform (hereinafter referred to as IDCT) is used as an orthogonal transform. ) Is done. At present, DCT / IDCT is realized by performing one-dimensional DCT / IDCT twice in the horizontal and vertical directions [Non-Patent Document 1]. DCT has a conversion efficiency equivalent to that of the Karhunen-Reeve transform (hereinafter referred to as KLT) for a model in which the correlation coefficient of the prediction error is regularly reduced according to the inter-pixel distance. The DCT in the horizontal and vertical directions is not always optimal depending on the edge component, etc., and research on changing the direction of applying the DCT according to the feature of the image [Non-Patent Document 2], DCT and discrete sine transform (Discrete Sine Transform, The following studies have been conducted to switch DST) [Patent Document 1].
A. Hallapuro, M. Karczewicz, and H. Malvar, “Low complexity transform and quantization,” Joint Video Team of ISO / IEC MPEG and ITU-T VCEG, Jan. 2002, JVT-B038 and JVT-B039. J. Fu, B. Zeng, “Diagonal Discrete Cosine Transforms for Image Coding”, PCM 2006 JP-A-11-55678

  H. The horizontal and vertical DCT of Non-Patent Document 1 used in H.264 provides conversion efficiency equivalent to that of KLT for a model in which the correlation coefficient of the prediction error is regularly reduced according to the inter-pixel distance. When an edge component is generated in the error signal, the DCT in the horizontal and vertical directions is not necessarily optimal, and there is a possibility that sufficient conversion efficiency cannot be obtained.

  Further, in Patent Document 1, conversion is performed by switching between DCT and DST depending on the feature of the image. DST has high conversion efficiency for pixels predicted to be interpolated. Since intra prediction in H.264 is extrapolation prediction, improvement in conversion efficiency cannot be expected even if DST is applied to a prediction error caused by intra prediction.

  In Non-Patent Document 2, the direction in which one-dimensional DCT is applied to reduce side information is the same as that of conventional H.264. In addition to the H.264 conversion method, the orthogonal right lower direction and the orthogonal left lower direction are limited to two directions, and the optimum conversion direction in terms of rate distortion is selected from these directions. However, since the direction of intra prediction and the directionality of the prediction error are similar, and the power of the coefficient is not concentrated even if the one-dimensional DCT is applied in a direction different from the prediction direction, each of the orthogonal lower right direction and the orthogonal lower left direction It is redundant to perform conversion in the direction, and side information for identifying each conversion direction leads to a decrease in encoding efficiency. Further, the base length when the one-dimensional DCT is applied obliquely is extremely short at the end of the coding block, and the effect of the conversion is not sufficiently exhibited.

  One aspect of the present invention is a moving image encoding method for encoding an input image signal, which generates a prediction error signal indicating a difference value between a prediction signal generated according to a predetermined prediction mode and the input image signal. The prediction error signal is generated by the orthogonal transformation of the prediction error signal by the first conversion method depending on the prediction direction of the prediction mode and the second conversion method not depending on the prediction direction to generate a conversion coefficient. An orthogonal transform step, a quantization step for performing a quantization process on the transform coefficient to generate a quantized transform coefficient, the quantized transform coefficient, and the first transform method or the second transform method. And a coding process step for generating encoded data by performing entropy coding processing on the conversion information indicating the above.

  H. using spatial correlation When intra prediction such as H.264 is performed, there is a high possibility that the prediction direction and the directionality of the prediction error are the same, so the power concentration does not decrease even if one-dimensional DCT in directions other than the prediction direction is not considered. Further, the system of the aspect of the present invention is a conventional H.264 standard. In addition to the H.264 conversion method, one candidate for the direction in which the one-dimensional DCT is applied is determined, so that side information can be reduced as compared with Non-Patent Document 2. Further, since the one-dimensional DCT having a long base length is applied to the pixels at the edge of the image by folding the pixels, the power concentration level is increased. From these three points, it is possible to finally reduce the code amount by performing quantization processing and entropy coding processing on the transform coefficient generated by this transform method. Conventionally, one-dimensional DCT is applied in two directions, horizontal and vertical. However, in this method, one-dimensional DCT is sufficient in the prediction direction, so that encoding and decoding processes are reduced.

  DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Exemplary embodiments of a moving picture coding method, a moving picture coding apparatus, a moving picture decoding method, and a moving picture decoding apparatus according to the present invention will be described below in detail with reference to the accompanying drawings.

[First embodiment]
(About the encoder)
A video encoding apparatus 100 according to the first embodiment shown in FIG. 1 includes a control unit 107 that controls conversion information from prediction mode information, a conversion method selection switch 101 that selects a conversion method based on the conversion information, Transform quantization units 102 to 105 that perform transform quantization corresponding to directions, and an entropy encoding unit 106 that encodes transform coefficients and transform information are provided.

  In FIG. 1, an input moving image signal 115 is divided for each macroblock or macroblock pair and input to the moving image encoding apparatus 100. In the video encoding device 100, the prediction unit 114 performs intra-frame prediction on the input video signal 115. The prediction unit 114 will be specifically described below.

  The prediction unit 114 performs prediction in units of sub blocks obtained by further dividing the macro block. For example, H.M. In the luminance signal in the H.264 intra prediction, there are a macroblock having 16 4 × 4 pixel subblocks, a macroblock having four 8 × 8 pixel subblocks, and a macroblock having one 16 × 16 pixel subblock. . FIG. 2A shows the relationship between the encoding target frame and the encoding target macroblock, FIG. 2B shows the macroblock having 4 × 4 pixel subblocks, and FIG. 2C shows the macroblock having 8 × 8 pixel subblocks. 2) shows a macroblock having 16 × 16 pixel sub-blocks in FIG. Nine prediction modes exist in the 4x4 subblock and the 8x8 subblock, and four prediction modes exist in the 16x16 subblock, and intra prediction is performed in each subblock.

  FIG. All prediction modes in 4 × 4 sub-block and 8 × 8 sub-block of H.264 intra prediction, that is, vertical prediction mode, horizontal prediction mode, DC prediction mode, orthogonal lower left prediction mode, orthogonal lower right prediction mode, vertical right prediction mode, horizontal lower prediction mode , Vertical left prediction mode, and horizontal top prediction mode. The prediction direction of each prediction mode is shown in FIG. A specific prediction method for the prediction direction shown in FIG. 4A will be described with reference to FIGS.

For the sub-block to be encoded in FIG. When the mode 2 DC prediction in the H.264 intra prediction mode is selected, the prediction pixel is calculated by Expression (1).
H = (A + B + C + D), V = (I + J + K + L)
a ~ p = (H + V + 4) >> 3
... (1).

  When the reference pixel cannot be used, a predicted value is generated using an average value of the available reference pixels. If there is no usable pixel, a prediction value is generated with a value that is half the maximum luminance value of the encoding device (128 for 8 bits). When the other mode is selected, the direction prediction unit 114 uses a prediction method that copies the prediction value interpolated from the reference pixel with respect to the prediction direction shown in FIG. Specifically, a prediction value generation method when mode 0 (vertical prediction) is selected will be described.

The following modes can be selected only when reference pixels A to D are available.
a, e, i, m = A
b, f, j, n = B
c, g, k, o = C
The details of the d, h, l, p = D prediction method are shown in FIG. 4C, and the luminance values of the reference pixels A to D are copied as they are in the vertical direction and supplemented as prediction values.

On the other hand, the prediction method when mode 4 (orthogonal lower right prediction) is selected will also be described using equation (2).
d = (B + (C << 1) + D + 2) >> 2
c, h = (A + (B << 1) + C + 2) >> 2
b, g, l = (M + (A << 1) + B + 2) >> 2
a, f, k, p = (I + (M << 1) + A + 2) >> 2
e, j, o = (J + (I << 1) + M +2) >> 2
i, n = (K + (J << 1) + I + 2) >> 2
m = (L + (K << 1) + J + 2) >> 2
... (2)
This prediction mode can be selected only when reference pixels A to D and I to M are available. The details of the main prediction mode are shown in FIG. 4D, and according to the main prediction mode, the value generated by the 3-tap filter is copied in the lower right 45 degrees direction and compensated as a predicted value.

  An almost similar framework is used for prediction methods other than prediction modes 0, 2, and 4, and an interpolation value is generated from reference pixels that can be used in the prediction direction, and the value is copied according to the prediction direction. Make a prediction. Moreover, although the prediction method was demonstrated about the 4x4 pixel block, it predicts by producing | generating the interpolation value from a reference pixel also about an 8x8 pixel block, and copying the value according to a prediction direction.

  As described above, the prediction error signal 116 is generated by taking the difference between the prediction image signal 121 generated by the prediction unit 114 and the input image signal 115. At this time, the encoding control unit 107 generates conversion information 118 according to the prediction mode information 120 selected by the prediction unit 114. The conversion information 118 is information for determining whether to perform DCT in the horizontal / vertical direction for orthogonal transform or to perform one-dimensional DCT along the prediction direction indicated by the prediction mode information. Used to switch the method selection switch 108. In the embodiment, the encoding control unit 107 outputs only horizontal / vertical conversion information when the prediction direction indicated by the prediction mode information is horizontal or vertical, and outputs horizontal / vertical conversion information and diagonal conversion information when the prediction direction is the diagonal direction. .

  The prediction error signal 116 is input to the conversion method selection switch 101, the switch is switched according to the conversion information 118, and is input to either the first conversion unit 102 or the second conversion unit 103. In the first conversion unit 102, the prediction error signal 116 is subjected to horizontal and vertical DCT by a horizontal and vertical DCT unit 200 as shown in FIG. The transform coefficient 232 from the horizontal / vertical DCT unit 200 is then quantized by a quantizer 201 corresponding to the horizontal / vertical DCT unit 200 in the first quantizing unit 104 as shown in FIG.

  In the second conversion unit 103, as shown in FIG. 6A, the prediction error signal 116 is DCT-transformed by the DCT unit selected from the DCT units 203 to 208 and corresponding to the prediction direction of the prediction mode information 120. . Thereafter, in the second quantization unit 105, as shown in FIG. 6B, a quantizer corresponding to the prediction mode information 120 is selected from the quantizers 210 to 215, and the transform coefficient of the second transform unit 103 is selected. 233 is quantized and a quantized transform coefficient 117 is generated.

Here, from the viewpoint of encoding efficiency, mode determination for determining an optimal mode will be described in detail. For all the prediction modes that can be taken by the sub-block, the encoding process after the orthogonal transformation indicated by the transformation information 118 is performed, and the mode that results in the lowest coding cost is determined as the optimum mode of the sub-block. . More specifically, the difference error SAD obtained by calculating the absolute value of the difference value between the prediction signal generated by the prediction method defined in the prediction mode and the original image signal and the side required when using the prediction mode The mode is determined from the information value OH using Equation (3).
K = SAD + λ × OH (3)
λ is given as a constant and is determined based on the value of the quantization scale. The mode is determined based on the cost K thus obtained. Generally, the mode that gives the smallest value of cost K is selected as the optimum mode.

  In this embodiment, the side information and the absolute value of the difference value are used. However, as another embodiment, the mode may be determined using only the mode information or only the absolute sum of the prediction error signals. May be subjected to Hadamard transform or an approximate value may be used. Further, the cost may be created using the activity of the input image, or the cost function may be created using the quantization scale.

As another embodiment for calculating cost, provisional encoding is performed, and the amount of code when the difference value between the prediction signal generated in the prediction mode of provisional encoding and the original image is actually encoded, and The mode may be determined using a square error between the locally decoded image obtained by locally decoding the transform coefficient and the input image signal. The mode determination formula in this case is as shown in Formula (4).
J = D + λ × R (4)
Here, D is an encoding distortion representing a square error between the input image and the locally decoded image. On the other hand, R represents a code amount estimated by provisional encoding. When this cost is used, entropy coding and local decoding (including inverse quantization and inverse transform processing) are required for each coding mode, so the circuit scale increases, but the exact code amount and coding Distortion can be used, and encoding efficiency can be kept high. For this cost, the cost may be calculated using only the code amount or only the coding distortion, or a cost function may be created using a value approximating these.

Next, a one-dimensional DCT method along the prediction direction will be described. First, H. The DCT performed in H.264 will be described. For a 4 × 4 sub-block, if the input signal is X, the transformation matrix is T, and the transformation coefficient is C, the transformation coefficient C is obtained according to Equation (5).

That is, the input image signal is subjected to the one-dimensional DCT in the horizontal direction using the transformation matrix T, and then subjected to the one-dimensional DCT in the vertical direction using the transposed matrix T ^t of the transformation matrix T.

  The above calculation is the same for the 8 × 8 sub-block. On the other hand, in the present embodiment, H.264 is applied to 4 × 4 subblocks and 8 × 8 subblocks. A DCT transform with directionality along the prediction direction is performed for six directions from among the nine types of intra prediction defined in H.264 except for three types of vertical prediction, horizontal prediction, and DC prediction.

First, H. A conversion method corresponding to each of the H.264 intra prediction modes 3 to 8 will be described using 4 × 4 sub-blocks illustrated in FIG. 7. In the prediction mode 3, the pixels in the sub-block are moved along the prediction direction (a), (b, e), (c, f, i), (d, g, j, m), (h, Pixels are grouped into 7 sets of k, n), (l, o), and (p). FIG. 8A shows a combination of pixel columns to which one-dimensional DCT is applied. Assuming that the conversion coefficient is F (u), a one-dimensional DCT with a base length of 4 is performed on the pixel column (d, g, j, m) corresponding to the diagonal line of the 4 × 4 pixel block as shown in Equation (6).

 (However, f (0) = d, f (1) = g, f (2) = j, f (3) = m).

Next, for the pixel column (c, f, i), a one-dimensional DCT with a base length of 3 is performed as in Equation (7).

 (However, f (0) = c, f (1) = f, f (2) = i).

Further, for the pixel columns (a) and (b, e), one-dimensional DCT with a base length of 3 is performed as shown in Equation (8) by folding and combining the pixels.

 (However, f (0) = b, f (1) = e, f (2) = a or f (0) = e, f (1) = b, f (2) = a).

Further, as another example of 4 × 4 conversion, a method of performing primary DCT without returning pixels is conceivable. FIG. 8B shows a combination of pixel columns to which the one-dimensional DCT is applied at this time. The pixel columns (d, g, j, m) and (e, f, i) are the same as those of the one-dimensional DCT, but the pixels are not folded and combined with respect to (b, e) and (a). On the other hand, one-dimensional DCT is performed as shown in Equation (9) and Equation (10).

(However, f (0) = b, f (1) = e)

 (However, f (0) = a).

As another example, a method of performing one-dimensional DCT by folding and joining all the pixel columns (c, f, i), (e, b), and (a) is conceivable. FIG. 8C shows a combination of pixel columns to which one-dimensional DCT is applied at this time. The pixel column (d, g, j, m) is the same as the one-dimensional DCT, but the pixel column (c, f, i), (e, b), (a) As such, one-dimensional DCT is performed.

(However, f (0) = c, f (1) = f, f (2) = i, f (3) = e, f (4) = b, f (5) = a
Or f (0) = i, f (1) = f, f (2) = c, f (3) = b, f (4) = e, f (5) = a)
For the remaining pixel columns (h, k, n), (l, e), (p), pixels that are diagonal lines of (e, f, i), (b, e), (a) and 4x4 blocks Since the line (d, g, j, m) is line symmetric, one-dimensional DCT is performed by the same calculation as (e, f, i), (b, e), and (a).

In the case of the prediction mode 4, since it is considered that the prediction direction of the prediction mode 3 is reversed left and right, the DCT calculation can be performed in the same manner as in the prediction mode 3.
In the above description, real DCT calculation is used. It is also possible to perform DCT by integer arithmetic as defined in H.264.

Next, DCT calculation methods corresponding to prediction modes 5 to 8 will be described.
In the case of the prediction mode 5, the pixels in the sub-block are moved along the prediction direction (a, e, j, n), (b, f, k, o), (c, g, l, p), ( Pixels are grouped into five groups i, m) and (d, h). FIG. 9A shows a combination of pixel columns to which one-dimensional DCT is applied. Assuming that the conversion coefficient is F (u), a one-dimensional DCT with a base length of 4 is performed on the pixel array (a, e, j, n) as shown in Equation (12).

(However, f (0) = a, f (1) = e, f (2) = j, f (3) = n)
The same calculation is performed on the pixel columns (b, f, k, o) and (c, g, l, p).

Next, for the pixel column (i, m), a one-dimensional DCT with a base length of 3 is performed as in Equation (13).

(However, f (0) = i, f (1) = m)
The same calculation as Expression (13) is performed on the pixel column (d, h).

Further, as another example of the DCT calculation corresponding to the prediction mode 5, a method of performing primary DCT by folding pixels together is conceivable. FIG. 9B shows a combination of pixel columns to which the one-dimensional DCT is applied at this time. The pixel column (b, f, k, o) is the same as the one-dimensional DCT, but (a, e, j, n), (m, i) and (p, l, g, c), ( With respect to d, h), the pixels are folded and combined to perform one-dimensional DCT as shown in Equation (14).

(However, f (0) = a, f (1) = e, f (2) = j, f (3) = n, f (4) = m, f (5) = i)
The same calculation as Expression (14) is performed for the pixel columns (p, l, g, c) and (d, h).

In the case of the prediction modes 6 to 8, since it is considered that the prediction direction of the prediction mode 5 is rotated or inverted, it is possible to perform the DCT calculation in the same manner as in the prediction mode 5.
In the above description, real DCT calculation is used. It is also possible to perform DCT by integer arithmetic as defined in H.264.

Next, for the 8 × 8 sub-block, Each of the H.264 intra prediction modes 3 to 8 will be described using an 8 × 8 block shown in FIG. 10.
In the prediction mode 3, the pixels in the sub-block are moved along the prediction direction (a0), (a1, b0), (a2, b1, c0), (a3, b2, c1, d0), (a4, b3, c2, d1, e0), (a5, b4, c3, d2, e1, f0), (a6, b5, c4, d3, e2, f1, g0), (a7, b6, c5, d4, e3, f2, g1, h0), (b7, c6, d5, e4, f3, g2, h1), (c7, d6, e5, f4, g3, h2), (d7, e6, f5, g4, h3), ( e7, f6, g5, h4), (f7, g6, h5), (g7, h6), and 15 groups of (h7). FIG. 11A shows a combination of pixel columns to which one-dimensional DCT is applied. Assuming that the conversion coefficient is F (u), the pixel length (a7, b6, c5, d4, e3, f2, g1, h0) corresponding to the diagonal line of the 8 × 8 pixel block is 1 with a base length of 8 as shown in Equation (15). Dimensional DCT is performed.

 (However, f (0) = a7, f (1) = b6, f (2) = c5, f (3) = d4, f (4) = e3, f (5) = f2, f (6) = g1, f (7) = h0).

Next, with respect to the pixel columns (a5, b4, c3, d2, e1, f0) and (a6, b5, c4, d3, e2, f1, g0), the pixel columns are folded back to obtain Equation (16) A one-dimensional DCT with a base length of 13 is performed.

(However, f (0) = a6,, f (1) = b5, f (2) = c4, f (3) = d3, f (4) = e2, f (5) = f1, f (6) = g0, f (7) = f0,
f (8) = e1,, f (9) = d2, f (10) = c3, f (11) = b4, f (12) = a5,
Or f (0) = g0,, f (1) = f1, f (2) = e2, f (3) = d3, f (4) = c4, f (5) = b5, f (6) = a6, f (7) = a5,
f (8) = b4,, f (9) = c3, f (10) = d2, f (11) = e1, f (12) = f0).

Further, for the pixel columns (a3, b2, c1, d0) and (a4, b3, c2, d1, e0), the pixel columns are folded and a one-dimensional DCT having a base length of 9 is performed as shown in Equation (17). Is called.

(However, f (0) = a4,, f (1) = b3, f (2) = c2, f (3) = d1, f (4) = e0, f (5) = d0, f (6) = c1, f (7) = b2,
f (8) = a3 or f (0) = e0,, f (1) = d1, f (2) = c2, f (3) = b3, f (4) = a4, f (5) = a3 , f (6) = b2, f (11) = c1, f (12) = d0).

Further, for the pixel columns (a0), (a1, b0), and (a2, b1, c0), the pixel columns are folded and combined, and a one-dimensional DCT with a base length of 6 is performed as shown in Equation (18).

(However, f (0) = a2, f (1) = b1, f (2) = c0, f (3) = b0, f (4) = a (1), f (5) = a (0)
Or, f (0) = c0, f (1) = b1, f (2) = a2, f (3) = a1, f (4) = b0, f (5) = a (0)).

  In the above, the pixel columns (a0), (a1, b0), (a2, b1, c0), (a3, b2, c1, d0), (a4, b3, c2, d1, e0), (a5, b4, c3, d2, e1, f0) and (a6, b5, c4, d3, e2, f1, g0) were subjected to one-dimensional DCT by combining adjacent pixel columns. , c5, d4, e3, f2, g1, h0), it is also possible to perform a one-dimensional DCT without combining the respective pixel columns. FIG. 11B shows a combination of pixel columns to which the one-dimensional DCT is applied at this time.

  (B7, c6, d5, e4, f3, g2, h1), (c7, d6, e5, f4, g3, h2), (d7, e6, f5, g4, h3), (e7, For (f6, g5, h4), (f7, g6, h5), (g7, h6), (a0), (a1, b0), (a2, b1, c0), (a3, b2, c1, d0) , (A4, b3, c2, d1, e0), (a5, b4, c3, d2, e1, f0), (a6, b5, c4, d3, e2, f1, g0) and a diagonal line of 8 × 8 blocks Since it is axisymmetric with respect to the pixel column (a7, b6, c5, d4, e3, f2, g1, h0), (a0), (a1, b0), (a2, b1, c0), (a3, b2, c1 , d0), (a4, b3, c2, d1, e0), (a5, b4, c3, d2, e1, f0), (a6, b5, c4, d3, e2, f1, g0) Perform one-dimensional DCT.

In the case of the prediction mode 4, since it is considered that the prediction direction of the prediction mode 3 is rotated or reversed, the DCT calculation can be performed in the same manner as in the prediction mode 3.
In the above description, real DCT calculation is used. It is also possible to perform DCT by integer arithmetic as defined in H.264.

  Next, DCT calculation methods corresponding to prediction modes 5 to 8 will be described.

In the case of the prediction mode 5, the pixels in the sub-block are (a0, b0, c1, d1, e2, f2, g3, h3), (a1, b1, c2, d2, e3, f3, g4, h4), (a2, b2, c3, d3, e4, f4, g5, h5), (a3, b3, c4, d4, e5, f5, g6, h6), (a4, b4, c5, d5, e6, f6, g7, h7), (c0, d0, e1, f1, g2, h2), (a5, b5, c6, d6, e7, f7), (e0, f0, g1, h1), (h0, g0), (a6, b6, c7, d7) and (a7, b7) are grouped into 11 groups of pixels. FIG. 12A shows a combination of pixel columns to which one-dimensional DCT is applied at this time. Assuming that the conversion coefficient is F (u), a one-dimensional DCT with a base length of 8 is performed on the pixel column (a0, b0, c1, d1, e2, f2, g3, h3) as shown in Equation (19).

(However, f (0) = a0, f (1) = b0, f (2) = c1, f (3) = d1, f (4) = e2, f (5) = f2, f (6) = g3, f (7) = h3)
Pixel column (a1, b1, c2, d2, e3, f3, g4, h4), (a2, b2, c3, d3, e4, f4, g5, h5), (a3, b3, c4, d4, e5, f5 , g6, h6) and (a4, b4, c5, d5, e6, f6, g7, h7), the same calculation is performed.

Next, for the pixel column (c0, d0, e1, f1, g2, h2), a one-dimensional DCT with a base length of 6 is performed as in Equation (20).

(However, f (0) = c0, f (1) = d0, f (2) = e1, f (3) = f1, f (4) = g2, f (5) = h2)
The same calculation is performed on the pixel columns (a5, b5, c6, d6, e7, f7).

Next, with respect to the pixel columns (e0, f0, g1, h1) and (h0, g0), the pixel columns are folded and combined, and a one-dimensional DCT with a base length of 6 is performed as shown in Equation (21).

(However, f (0) = e0, f (1) = f0, f (2) = g1, f (3) = h1, f (4) = h0, f (5) = g0)
The same calculation is performed on the pixels (a6, b6, c7, d7) and (a7, b7).

  In the above, the pixel columns (e0, f0, g1, h1), (h0, g0), (d7, c7, b6, a6), (a7, b7) are combined with adjacent pixel columns. The one-dimensional DCT is performed, but the one-dimensional DCT is performed without combining the respective pixel columns, similarly to the one-dimensional DCT performed on (a0, b0, c1, d1, e2, f2, g3, h3). It is also possible. FIG. 12B shows a combination of pixel columns to which the one-dimensional DCT is applied at this time.

In the case of the prediction modes 6 to 8, since it is considered that the prediction direction of the prediction mode 5 is rotated or reversed, it is possible to perform the DCT calculation as in the case of the prediction mode 5.
In the above description, real DCT calculation is used. It is also possible to perform DCT by integer arithmetic as defined in H.264.

  In this embodiment, the 4 × 4 size and the 8 × 8 size are used as the block size of the sub-block to be encoded. However, as another embodiment, a block having a size smaller than the 4 × 4 size or a size larger than the 8 × 8 size is used. It is also possible to perform the encoding process using these blocks as sub-blocks. For example, the encoding process may be performed with a block of 2 × 2 size or 16 × 16 size.

  In this embodiment, DCT is used as the orthogonal transformation method. However, as another embodiment, KLT, DST, and discrete wavelet transform (hereinafter referred to as DWT) can be used.

The entropy encoding unit 106 includes a quantized transform coefficient 117 generated by the first transform unit 102 and the first quantization unit 104 or the second transform unit 103 and the second quantization unit 105, and an encoding control unit 107. The generated conversion information 118 is input and entropy coding is performed. The conversion information 118 and the conversion coefficient 117 of the mode determined to be optimal by the mode determination are entropy encoded, and encoded data 124 is generated.
In addition, when the prediction mode determined as the optimum mode by the mode determination is average prediction, prediction in the vertical direction or horizontal direction, the entropy encoding unit 106 does not encode the conversion information 118, and only the quantized conversion coefficient 117 is used. Entropy encoding.

  The quantized transform coefficient 117 is input to the inverse transform method selection switch 108. The inverse transformation method selection switch 108 is switched according to the value of the transformation information 118, and the quantized transformation coefficient 117 is input to the second inverse quantization unit 109 or the first inverse quantization unit 110.

In the first inverse quantization unit 110, as shown in FIG. 13A, inverse quantization corresponding to inverse DCT in the horizontal and vertical directions is performed on the quantized transform coefficient 117 by the inverse quantizer 216, and the transform coefficient is Thereafter, as shown in FIG. 13B, the inverse DCT in the horizontal and vertical directions is performed by the inverse DCT unit 217 in the first inverse transform unit 112, and the decoded prediction error signal 119 is generated. In addition, in the second inverse quantization unit 109, as shown in FIG. 14A, the quantization transform coefficient 117 is subjected to inverse quantization corresponding to inverse DCT in the prediction direction of the prediction mode information 120 by an inverse quantizer (219 to 224). ). Further, as shown in FIG. 14B, the transform coefficient is obtained by the inverse DCT by the inverse DCT unit corresponding to the prediction direction of the prediction mode information 120 selected from the inverse DCT units 226 to 231 by the second inverse transform unit 111. And a decoded prediction error signal 119 is generated.
The decoded prediction error signal 119 is added to the prediction image signal 121 generated by the prediction unit 114, stored in the reference memory 113, and then input to the prediction unit 114 as the reference image signal 122.

  The above is the configuration of the video encoding apparatus 100 according to the present embodiment. Hereinafter, the moving picture coding method according to the present embodiment will be described with reference to the flowchart of FIG.

  When a moving image signal is input to the moving image encoding device 100 for each frame, encoding of the input image is started for each macroblock or macroblock pair (step S301). In encoding the macroblock, the subblock index is initialized to sub_blk = 0, and the number of subblocks MAX_SUB_BLK is set at the same time (step S302).

  It is determined whether or not the sub-block index sub_blk is smaller than MAX_SUB_BLK, and it is confirmed that there is an uncoded sub-block in the sub-block in the macro block (step S303). If the determination in step S303 is “Yes”, the prediction mode index is initialized to mode_idx = 0, and the number of modes MAX_MODE that can be selected is set (step S304).

  Next, it is determined whether or not mode_idx is smaller than MAX_MODE (step S305). If the determination in step S305 is “Yes”, trans_idx is set according to the value of mode_idx. That is, when mode_idx is a mode number indicating a prediction mode in an oblique direction, trans_idx = 0 is set, and when mode_idx is a mode number indicating a prediction mode not in an oblique direction, trans_idx is not generated (step S306). . Further, the prediction unit 114 generates a prediction image signal 121 in the prediction mode mode_idx (step S307). By taking the difference between the predicted image signal 121 generated by the prediction unit 114 and the input image signal 115, the prediction error signal 116 in the prediction mode mode_idx is obtained (step S308).

  Next, the encoding control unit 107 determines whether the prediction mode mode_idx selected by the prediction unit 114 is smaller than mode_direction which is the smallest value among the mode numbers indicating the prediction modes in the oblique direction, or trans_idx == 1 It is determined whether or not (step S309). If either of the two conditions in the determination in step S309 is satisfied, the conversion method selection switch 101 is switched to the first conversion unit 102, and the horizontal / vertical DCT is performed on the prediction error signal 116 (step S310). The coefficient is quantized corresponding to horizontal and vertical DCT, and a quantized transform coefficient 117 is generated (step S312).

  When neither of the two conditions in the determination in step S309 is satisfied, the selection switch 101 is switched to the second conversion unit 103, and the switch 202 of the second conversion unit 103 switches the converters 203 to 208 according to the prediction mode mode_idx. DCT corresponding to the prediction direction is performed on the prediction error signal 116 by the converter selected by the switch 202 (step 311). Thereafter, the transform coefficient is quantized corresponding to each DCT in the second quantizing unit 105 by the corresponding quantizers of the quantizers 210 to 215 of the second quantizing unit 105, and the quantized transform coefficient 117 is obtained. It is generated (step S313).

  That is, only when the prediction direction is not the vertical direction or the horizontal direction, the conversion method between the vertical horizontal direction and the prediction direction conversion method according to the prediction mode is switched. When the prediction direction is the vertical direction and the horizontal direction, the horizontal direction is always vertical. Select the direction conversion method. In short, when the prediction direction is an oblique direction, conversion in the vertical and horizontal directions and conversion in the oblique direction are performed, and when the prediction directions are the vertical direction and the horizontal direction, only conversion in the vertical and horizontal directions is performed.

  Next, the quantized transform coefficient 117 and the transform information 118 are encoded by the entropy encoding unit 106 (step S314). At this time, if mode_idx <mode_direction, the conversion information trans_idx is not generated, so the conversion information 118 is not encoded, and only the conversion coefficient 117 is encoded. That is, when encoding the transform information and the transform coefficient, the entropy coding unit 106 codes the transform information in addition to the transform coefficient only when the prediction direction indicated by the prediction mode information is not the vertical direction and the horizontal direction. When the prediction directions indicated by the prediction mode are the vertical direction and the horizontal direction, the conversion information is not encoded, and only the conversion coefficient is encoded.

  After entropy coding in step S314, the prediction mode RD cost is stored (step S315), and then it is determined whether or not trans_idx is 1 (step S316). In step S316, if trans_idx == 0, 1 is added to trans_idx (step S317), the process returns to step S309, and branch determination is performed again based on the prediction mode mode_idx and the conversion information trans_idx.

  After steps S309 to S315, it is determined again whether or not trans_idx is 1 in step S316. If it is determined in step S316 that trans_idx == 1, 1 is added to mode_idx, and the next prediction mode is set (step S318). Thereafter, it is confirmed whether or not the set prediction mode mode_idx is smaller than MAX_MODE (step S305). If it is smaller, the processes of steps S306 to S318 are performed. After performing the processing from step S306 to step S318 for all possible modes, if the determination of mode_idx <MAX_MODE is “NO” in step S305, 1 is added to the subblock index sub_blk, The process proceeds to the encoding process for the next sub-block (step S319). It is determined whether or not the next sub-block is smaller than MAX_SUB_BLK (step S303). If the determination in step S303 is “Yes”, the processing from step S304 to step S319 is performed. If the determination of sub_blk <MAX_SUB_BLK is “NO” after steps S304 to S319 have been completed for all the subblocks, the cost functions of the respective modes obtained for the respective subblocks are compared, and the optimum mode of each block is compared. Is loaded (step S320), and the encoded data when encoding is performed in the optimum mode is multiplexed with a macroblock and transmitted (step S321).

Next, the syntax used in this embodiment will be described. The overall structure of the syntax is shown in FIG. The syntax used in the present embodiment is mainly composed of three parts, and High Level Syntax is packed with syntax information of an upper layer higher than a slice. Slice Level Syntax specifies information required for each slice, and Macroblock Level Syntax specifies error signal and mode information encoded in variable length required for each macroblock.
Each syntax is composed of more detailed syntax, and in High Level Syntax, it is composed of sequences such as Sequence parameter set syntax and Picture parameter set syntax, and syntax of picture level. Slice Level Syntax consists of Slice header syntax, Slice data syntax, etc. Furthermore, Macroblock Level Syntax is composed of Macroblock layer syntax, Macroblock prediction syntax, and the like.

  In the present embodiment, required syntax information is a sequence header, a slice header, and a macroblock header, and each syntax will be described below.

  The directional_dct_seq_flag shown in the sequence header of FIG. 19 is a flag indicating whether or not the orthogonal transform method of the present embodiment is used in the encoding target sequence. When this flag is 1, the directional_dct_seq_flag is set in the encoding target sequence. It is possible to use a form of orthogonal transform scheme. Next, directional_dct_pic_flag shown in the picture header of FIG. 20 is a flag indicating whether or not the orthogonal transform method of the present embodiment is used for a picture. When this flag is 1, the orthogonal of the present embodiment is used for a picture. A conversion method can be used. Next, directional_dct_slice_flag shown in the slice header of FIG. 21 is a flag indicating whether or not the orthogonal transform method of the present embodiment is used in the slice. When this flag is 1, the orthogonal of the present embodiment is used in the slice. A conversion method can be used.

  As shown in FIG. 22, the macroblock header has two flags of directional_dct4x4_mode_flag and directional_dct8x8_mode_flag as syntax in the macroblock layer. In this embodiment, when a macroblock is an intra macroblock and any of directional_dct_seq_flag, directional_dct_pic_flag, and directional_dct_slice_flag is 1, if the subblock in the macroblock is 4x4 block, directional_dct_4x4_mode_flag is set and subblock is 8x8 block If there is, one bit of directional_dct8x8_mode_flag is generated for each macroblock. These two flags are transformation information indicating whether the orthogonal transformation is horizontal / vertical DCT or DCT depending on the prediction direction. When each flag indicates 1, horizontal / vertical DCT is used. Indicates that DCT is being performed, and a value of 0 indicates that DCT depending on the prediction direction is being performed.

  In this embodiment, for each macroblock, transform information indicating whether orthogonal transform is performing horizontal / vertical DCT or DCT depending on the prediction direction is determined. However, as another embodiment, A method of determining conversion information for each sub-block, or determining conversion information according to auxiliary information for each sub-block can be considered.

  As described above, according to the first embodiment, in addition to the conventional H.264 conversion method, one-dimensional DCT is applied to each block while limiting the conversion direction to the direction in which intra prediction is performed. In addition, when the one-dimensional DCT is applied in an oblique direction, the one-dimensional DCT having a long base length is applied by folding back pixels in consideration of the direction of intra prediction. In addition, when the one-dimensional DCT is applied in the prediction direction, the one-dimensional DCT is applied only once in the prediction direction. In other words, when the prediction direction is diagonal, the vertical horizontal conversion and the diagonal conversion of the prediction direction are performed, and the diagonal conversion is narrowed down to one type corresponding to the prediction direction, and the vertical horizontal conversion result and the prediction direction are matched. The optimum conversion method is adopted by comparing the result of conversion in the diagonal direction.

[Second Embodiment]
Next, a second embodiment will be described. Since the overall configuration of the moving picture coding apparatus according to the present embodiment is substantially the same as that of the first embodiment, only differences from the first embodiment will be described. In the first embodiment, whether or not the orthogonal transform method of the present embodiment is used is determined for each macroblock whether the orthogonal transform performs horizontal / vertical DCT or performs DCT depending on the prediction direction. In this embodiment, the conversion information is switched in units of sub-macroblocks. FIG. 23 shows the syntax structure in the macroblock header in the present embodiment.

  In this embodiment, when a macroblock is an intra macroblock and any of directional_dct_seq_flag, directional_dct_pic_flag, and directional_dct_slice_flag is 1, if the subblock in the macroblock is 4x4 block, directional_dct_4x4_mode_flag is set and subblock is 8x8 block If there is, one bit of directional_dct8x8_mode_flag is generated for each sub macroblock. That is, 16 directional_dct4x4_mode_flag exists in one macroblock when the macroblock performs 4x4 conversion, and 4 directional_dct8x8_mode_flag exists in one macroblock when the 8x8 conversion is performed.

[Third embodiment]
Next, a third embodiment of the present invention will be described. Since the overall configuration of the moving picture encoding apparatus in the present embodiment is substantially the same as that of the second embodiment, only differences from the second embodiment will be described. In this embodiment, as in the second embodiment, when the macroblock is an intra macroblock and any of directional_dct_seq_flag, directional_dct_pic_flag, and directional_dct_slice_flag is 1, the subblock in the macroblock is a 4 × 4 block. If there is, directional_dct_4x4_mode_flag is generated, and if the subblock is an 8x8 block, directional_dct8x8_mode_flag is generated for one bit per subblock. At this time, unlike the second embodiment, instead of creating a flag for all sub-blocks, using prediction mode information and auxiliary information of coded_block_pattern indicating whether the non-zero coefficient is included in the sub-block Determine whether to generate a flag in the sub-block. FIG. 24 shows the syntax structure in the macroblock header in the present embodiment. intra8x8_pred_mode and intra4x4_pred_mode are values representing prediction modes of the 8x8 block and the 4x4 block, respectively, and the relationship between the prediction direction and the value of the prediction mode is the same as the value of FIG. When encoding, when the prediction direction indicated by the prediction mode is the vertical direction, the horizontal direction, or the DC direction, and when it is determined that the 8 × 8 pixel block including the sub-block does not have a non-zero coefficient from Coded_Block_Pattern Since there is no need to perform conversion depending on the prediction direction, directional_dct4x4_mode_flag or directional_dct8x8_mode_flag is not generated.

[Fourth embodiment]
FIG. 16 is a block diagram illustrating a configuration of a video decoding device 400 according to the fourth embodiment. The video decoding apparatus 400 according to the present embodiment selects an entropy decoding unit 401 that analyzes transform coefficients, prediction mode information, and transform information from encoded data, and an inverse quantization and inverse transform method from the analyzed transform information. Inverse transform method selection switch 402, and inverse quantization transform units 403 to 406 that perform inverse quantization and inverse transform corresponding to the direction of prediction obtained from the prediction mode information.

  In FIG. 16, encoded data 409 sent from the moving image encoding apparatus 100 and sent via the transmission system or the storage system is separated on the basis of the syntax for each frame by a demultiplexer (not shown). Is then input to the entropy decoding unit 401. In the entropy decoding unit 401, the variable length code of each syntax of the encoded data is decoded, and the transform information 410, the prediction mode information 411, the quantized transform coefficient 412 and the like are reproduced.

  The quantized transform coefficient is input to the inverse transform method selection switch 402, and based on the transform information 410 decoded by the entropy decoding unit 401, either the first inverse quantizer 403 or the second inverse quantizer 404. Connected. When connected to the first inverse quantization unit 403, the quantized transform coefficients are inversely quantized by an inverse quantization method corresponding to horizontal and vertical DCT. On the other hand, when connected to the second inverse quantization unit 404, the inverse quantization method is selected from the inverse quantizers 219 to 224 by the prediction mode information 411 reproduced by the entropy decoding unit 401, and the prediction mode information indicates Inverse quantization corresponding to DCT in the prediction direction is performed. The transform coefficient inversely quantized by the first inverse quantization unit 403 or the second inverse quantization unit 404 is input to the corresponding first inverse transform unit 405 or second inverse transform unit 406. The first inverse transform unit 405 performs horizontal / vertical inverse DCT, and the second inverse transform unit 406 selects the inverse DCT method from the inverse DCT units 226 to 231 based on the prediction mode information 411, and indicates the prediction indicated by the prediction mode. The direction-dependent inverse DCT is performed, and a decoded prediction error signal 413 is output.

  A prediction image 415 is generated by the prediction unit 408 using the reference image 414 and the prediction mode information 411 stored in the reference memory 407. A decoded image signal is generated by adding the decoded prediction error signal 413 and the predicted image 415, accumulated in the reference memory 407, and then output as a reproduced image.

  In this embodiment, IDCT is used as the orthogonal inverse transform method. However, as another embodiment, inverse transform of KLT, DST, and DWT can be used.

  The above is the configuration of the moving picture decoding apparatus 400 according to the present embodiment. Hereinafter, the moving picture decoding method according to the present embodiment of the present invention will be described with reference to the flowchart of FIG. 17 by taking the moving picture decoding apparatus 400 as an example.

  When the moving image decoding apparatus 400 obtains moving image encoded data for each frame (step S501), decoding of the input moving image encoded data is started for each macroblock or each macroblock pair. The First, the encoded data is input to the entropy decoding unit 401, the variable length code of each syntax of the encoded data is decoded, and the transform information 410, the prediction mode information 411, the quantized transform coefficient 412 and the like are reproduced (step S502). ). Further, syntax analysis is performed from the data decoded in step S502 (step S503).

  Next, the sub-block index is initialized to sub_blk = 0 (step S504). It is determined whether or not sub_blk is smaller than the number of sub-blocks MAX_SUB_BLK of the macroblock (step S505). If smaller, prediction mode information mode_idx and conversion information trans_idx are loaded from the syntax analyzed in step S503 (step S506). Thereafter, the prediction unit 408 generates a prediction image of the prediction mode mode_idx loaded in step S506 (S507).

  Next, it is determined whether or not the prediction mode mode_idx is smaller than the smallest prediction mode number mode_direction indicating prediction in the oblique direction, or trans_idx == 1 (step S508). If the determination in step S508 is “Yes”, inverse quantization corresponding to horizontal and vertical DCT is performed (step S509), and then horizontal and vertical inverse DCT is performed (step S511). If the determination in step S508 is “NO”, inverse quantization corresponding to DCT in the intra prediction direction is performed (step S510), and then inverse DCT depending on the intra prediction direction is performed (step S512). Thereafter, the decoded error signal and the predicted image are combined (step S513) and stored in the reference memory 407 (step S514). Thereafter, 1 is added to the sub-block index sub_blk, and the process returns to step S505. If the determination in step S505 is “NO”, decoding of all the sub-blocks of the macroblock is completed and a decoded image is output (step S516).

  That is, when the prediction direction is diagonal, the inverse one of vertical and horizontal reverse conversion or diagonal prediction conversion is performed from the decoded conversion information. When the prediction direction is not diagonal, the vertical horizontal reverse conversion is always performed. I do.

Next, the syntax used in this embodiment will be described. The overall structure of the syntax is shown in FIG. The syntax used in the present embodiment is mainly composed of three parts, and High Level Syntax is packed with syntax information of an upper layer higher than a slice. Slice Level Syntax specifies information required for each slice, and Macroblock Level Syntax specifies error signal and mode information encoded in variable length required for each macroblock.
Each of these syntaxes is composed of more detailed syntax, and in High Level Syntax, it is composed of sequences such as Sequence parameter set syntax and Picture parameter set syntax, and picture level syntax. Slice Level Syntax consists of Slice header syntax, Slice data syntax, etc. Furthermore, Macroblock Level Syntax is composed of Macroblock layer syntax, Macroblock prediction syntax, and the like.

In the present embodiment, required syntax information is a sequence header, a slice header, and a macroblock header, and each syntax will be described below.
The directional_dct_seq_flag shown in the sequence header of FIG. 19 is a flag indicating whether or not the orthogonal transform method of the present embodiment is used in the encoding target sequence. When this flag is 1, the directional_dct_seq_flag is set in the encoding target sequence. It is possible to use a form of orthogonal transform scheme. Next, directional_dct_pic_flag shown in the picture header of FIG. 20 is a flag indicating whether or not the orthogonal transform method of the present embodiment is used for a picture. When this flag is 1, the orthogonal of the present embodiment is used for a picture. A conversion method can be used. Next, directional_dct_slice_flag shown in the slice header of FIG. 21 is a flag indicating whether or not the orthogonal transform method of the present embodiment is used in the slice. When this flag is 1, the orthogonal of the present embodiment is used in the slice. A conversion method can be used.

  As shown in FIG. 22, the macroblock header has two flags of directional_dct4x4_mode_flag and directional_dct8x8_mode_flag as syntax in the macroblock layer. In this embodiment, when a macroblock is an intra macroblock and any of directional_dct_seq_flag, directional_dct_pic_flag, and directional_dct_slice_flag is 1, if the subblock in the macroblock is 4x4 block, directional_dct_4x4_mode_flag is set and subblock is 8x8 block If there is, one bit of directional_dct8x8_mode_flag is generated for each macroblock. These two flags are transformation information indicating whether the orthogonal transformation is horizontal / vertical DCT or DCT depending on the prediction direction. When each flag indicates 1, the horizontal / vertical DCT is used. Indicates that DCT is being performed, and a value of 0 indicates that DCT depending on the prediction direction is being performed.

  In this embodiment, for each macroblock, transform information indicating whether orthogonal transform is performing horizontal / vertical DCT or DCT depending on the prediction direction is determined. However, as another embodiment, A method of determining conversion information for each sub-block or determining conversion information according to auxiliary information in units of sub-blocks can be considered.

[Fifth embodiment]
Next, a fifth embodiment of the present invention will be described. Since the overall configuration of the moving picture decoding apparatus in the present embodiment is substantially the same as that of the fourth embodiment, only differences from the fourth embodiment will be described.

  In the fourth embodiment, whether or not the orthogonal transform method of the present embodiment is used is determined for each macroblock whether the orthogonal transform performs horizontal / vertical DCT or performs DCT depending on the prediction direction. In this embodiment, the conversion information is switched in units of sub-macroblocks.

  FIG. 23 shows the syntax structure in the macroblock header in the present embodiment. In this embodiment, when a macroblock is an intra macroblock and any of directional_dct_seq_flag, directional_dct_pic_flag, and directional_dct_slice_flag is 1, if the subblock in the macroblock is 4x4 block, directional_dct_4x4_mode_flag is set and subblock is 8x8 block If there is, directional_dct8x8_mode_flag is generated by 1 bit per sub macroblock. That is, 16 directional_dct4x4_mode_flag exists in one macroblock when the macroblock performs 4x4 conversion, and 4 directional_dct8x8_mode_flag exists in one macroblock when the 8x8 conversion is performed.

[Sixth embodiment]
Next, a sixth embodiment of the present invention will be described. Since the overall configuration of the moving picture decoding apparatus in the present embodiment is almost the same as that of the fifth embodiment, only differences from the fifth embodiment will be described.

  In the present embodiment, as in the fifth embodiment, when the macroblock is an intra macroblock and any of directional_dct_seq_flag, directional_dct_pic_flag, and directional_dct_slice_flag is 1, the subblock in the macroblock is a 4 × 4 block. If there is, directional_dct_4x4_mode_flag is generated, and if the subblock is an 8x8 block, directional_dct8x8_mode_flag is generated for one bit per subblock. At this time, unlike the fifth embodiment, instead of creating a flag for all sub-blocks, using prediction mode information and auxiliary information of coded_block_pattern indicating whether the non-zero coefficient is included in the sub-block Determine whether to generate a flag in the sub-block.

  FIG. 24 shows the syntax structure in the macroblock header in the present embodiment. intra8x8_pred_mode and intra4x4_pred_mode are values representing prediction modes of the 8x8 block and the 4x4 block, respectively, and the relationship between the prediction direction and the value of the prediction mode is the same as the value of FIG. When encoding, when the prediction direction indicated by the prediction mode is the vertical direction, the horizontal direction, or the DC direction, and when it is determined that the 8 × 8 pixel block including the sub-block does not have a non-zero coefficient from Coded_Block_Pattern Since there is no need to perform conversion depending on the prediction direction, directional_dct4x4_mode_flag or directional_dct8x8_mode_flag is not generated. When there is no directional_dct4x4_mode_flag or directional_dct8x8_mode_flag in the decoded data, the decoder assumes that the value of directional_dct4x4_mode_flag or directional_dct8x8_mode_flag is 1, and performs subsequent decoding processing.

1 is a block diagram showing a configuration of an image encoding device according to a first embodiment. H. The figure which shows the relationship between the macroblock in H.264, and the subblock of each size. H. The figure which shows all the prediction modes of H.264 intra prediction. H. The figure which shows the prediction direction of H.264 intra prediction. The figure which shows the structure of the 1st conversion part concerning the 1st Embodiment, and a 1st quantization part. The figure which shows the structure of the 2nd conversion part concerning the same embodiment, and a 2nd quantization part. 4 × 4 pixel sub-block according to the embodiment. The figure which shows the one-dimensional DCT method (4x4 block, prediction mode 3) concerning the embodiment The figure which shows the one-dimensional DCT method (4x4 block, prediction mode 5) concerning the embodiment The figure which shows the 8x8 pixel subblock concerning the embodiment. The figure which shows the one-dimensional DCT method (8x8 block, prediction mode 3) concerning the embodiment The figure which shows the one-dimensional DCT method (8x8 block, prediction mode 5) concerning the embodiment The figure which shows the structure of the 1st inverse transformation part concerning the same embodiment, and a 1st inverse quantization part. The figure which shows the structure of the 2nd inverse transformation part concerning the same embodiment, and a 2nd inverse quantization part. The processing flowchart of the moving image encoding method by the image encoding apparatus. 1 is a block diagram showing a configuration of an image decoding device according to a first embodiment of the present invention. The processing flowchart of the moving image decoding method by the image decoding apparatus. The figure which shows the outline of the whole syntax structure concerning the embodiment. The figure which shows the outline of the data structure of the sequence header concerning 1st, 4th embodiment. The figure which shows the outline of the data structure of the picture header concerning 1st, 4th embodiment. The figure which shows the outline of the data structure of the slice header concerning 1st, 4th embodiment. The figure which shows the data structure outline | summary of the macroblock header concerning 1st, 4th embodiment. The figure which shows the data structure outline | summary of the macroblock header concerning 2nd, 5th embodiment. The figure which shows the data structure outline | summary of the macroblock header concerning 3rd, 6th embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Moving image encoding apparatus, 101 ... Conversion method selection switch, 102 ... 1st conversion part, 103 ... 2nd conversion part, 104 ... 1st quantization part, 105 ... 2nd quantization part, 106 ... Entropy encoding 107: Encoding control unit, 108: Inverse transform method selection switch, 109 ... Second inverse quantization unit, 110 ... First inverse quantization unit, 111 ... Second inverse transform unit, 112 ... First inverse transform unit , 113 ... Reference memory, 114 ... Prediction unit, 115 ... Input image signal, 116 ... Prediction error signal, 117 ... Quantization transform coefficient, 118 ... Conversion information, 119 ... Decoded prediction error signal, 120 ... Prediction mode information, 121 ... Predicted image signal, 122 ... reference image signal, 124 ... encoded data, 200 ... horizontal / vertical DCT, 201 ... quantizer for horizontal / vertical DCT, 202 ... switch, 203-208 ... DCT, 210-2 DESCRIPTION OF SYMBOLS 15 ... Quantizer, 216 ... Inverse quantizer for horizontal / vertical DCT, 217 ... Horizontal / vertical inverse DCT, 218 ... Switch, 219-224 ... Inverse quantizer, 225 ... Switch, 226-231 ... Inverse DCT, 401 ... entropy decoding unit, 402 ... inverse transformation method selection switch, 403 ... first inverse quantization unit, 404 ... second inverse quantization unit, 405 ... first inverse transformation unit, 406 ... second inverse transformation unit, 407 ... Reference memory, 408 ... prediction unit, 409 ... encoded data, 410 ... transform information, 411 ... prediction mode information, 412 ... quantized transform coefficient, 413 ... decoded prediction error signal, 414 ... reference image, 415 ... predicted image, 416 ... Decoded image signal, 417 ... Decoding control unit

Claims (16)

A moving image encoding method for encoding an input image signal,
A prediction error signal generation step for generating a prediction error signal indicating a difference value between a prediction signal generated according to a predetermined prediction mode and an input image signal;
An orthogonal transform step of orthogonally transforming the prediction error signal to generate transform coefficients by a first transform method that depends on a prediction direction of the prediction mode or a second transform method that does not depend on the prediction direction;
A quantization step of performing a quantization process on the transform coefficient to generate a quantized transform coefficient;
An encoding process step of performing entropy encoding processing on the quantized transform coefficient and transform information indicating the first transform method or the second transform method to generate encoded data;
A moving picture encoding method including: When the prediction direction is not the vertical direction or the horizontal direction, the first conversion method or the second conversion method is selected. When the prediction direction is the vertical direction or the horizontal direction, the first conversion method is selected. A conversion method selection step to select;
The moving image encoding method according to claim 1, wherein the orthogonal transformation step performs orthogonal transformation according to a selected transformation method.   The orthogonal transformation step, when the first transformation method is selected in the transformation method selection step, orthogonally transforms the prediction error signal along a prediction direction indicated by a prediction mode. The moving image encoding method according to Item 2.   In the orthogonal transforming step, when performing orthogonal transform along the prediction direction, one-dimensional orthogonal transform is performed on a pixel sequence obtained by folding back and joining adjacent spatially linear pixel sequences at a block end. The moving picture encoding method according to any one of claims 1 to 3.   The orthogonal transformation step, when performing orthogonal transformation along the prediction direction, orthogonal transformation is performed using any of DCT, KLT, DST, and DWT as a method of orthogonal transformation. 5. The moving image encoding method according to any one of 1 to 4.   In the orthogonal transform step, when the first transform method depending on the prediction direction is performed, the quantization step performs quantization using a different quantization method for each orthogonal transform according to the first orthogonal transform method. The moving picture coding method according to claim 1, wherein   The entropy encoding step encodes the conversion information in addition to the conversion coefficient only when the prediction direction indicated by the prediction mode information is not the vertical direction and the horizontal direction when encoding the conversion information and the conversion coefficient. The conversion information is not encoded when the prediction directions indicated by the modes are the vertical direction and the horizontal direction, and only the conversion coefficient is encoded. Video encoding method.   In the transform method selection step, the orthogonal transform method of the encoding target block is changed from the first transform method or the second transform method to at least one unit of sequence, picture, slice, or macroblock. The moving picture coding method according to claim 1, wherein the moving picture coding method is changed by: A moving image encoding device for encoding an input image signal,
A prediction error signal generation unit that generates a prediction error signal indicating a difference value between a prediction signal generated according to a predetermined prediction mode and an input image signal;
An orthogonal transform unit that orthogonally transforms the prediction error signal and generates transform coefficients by a first transform method that depends on a prediction direction of the prediction mode or a second transform method that does not depend on the prediction direction;
A quantization unit that performs a quantization process on the transform coefficient and generates a quantized transform coefficient;
An encoding unit that performs entropy encoding processing on the quantized conversion coefficient and conversion information indicating the first conversion method or the second conversion method, and generates encoded data;
A video encoding apparatus including: A decoding step of decoding the input encoded data according to a predetermined method and deriving orthogonal transform coefficients necessary for the decoding process, prediction mode information, and transform information;
Inverse transform method selection for selecting a first inverse transform method that depends on the prediction direction or a second inverse transform method that does not depend on the prediction direction as an inverse transform method of the decoding target block of the encoded data according to the transform information Steps,
An inverse quantization step of performing an inverse quantization process on the orthogonal transform coefficient obtained in the decoding step to generate an orthogonal transform coefficient;
According to the inverse transformation method selected in the inverse transformation method selection step, the orthogonal transformation coefficient is inversely orthogonal transformed to generate a decoded prediction error signal; and
Generating a prediction signal from the decoded decoded image signal based on the encoding mode information, and adding the prediction signal to the decoded prediction error signal to generate a decoded image;
A video decoding method comprising:   In the inverse transform selection step, when the first transform method is selected, the inverse quantization step performs inverse quantization with a different inverse quantization method for each orthogonal transform depending on the prediction direction. The moving picture decoding method according to claim 11, wherein:   11. The inverse transform selection step, when the first transform method is selected, the inverse orthogonal transform step performs an inverse orthogonal transform along a prediction direction indicated by a prediction mode. The moving picture decoding method according to claim 11.   In the inverse orthogonal transform step, when performing the inverse orthogonal transform along the prediction direction, the one-dimensional inverse orthogonal transform is not performed only on the spatially linear transform coefficients, but adjacent spatially straight lines are performed. 13. The moving picture decoding method according to claim 11, wherein one-dimensional inverse orthogonal transform is performed on a transform coefficient sequence obtained by folding transform transform coefficients at a block end and combining them.   The inverse orthogonal transform step performs inverse orthogonal transform using any one of inverse DCT, inverse KLT, inverse DST, and inverse DWT as an inverse orthogonal transform method when performing orthogonal transform along the prediction direction. The moving picture decoding method according to claim 11, wherein: the moving picture decoding method according to claim 11.   In the inverse transform method selection step, the inverse orthogonal transform method of the decoding target block of the encoded data is the first inverse transform method or the first inverse transform method in at least one unit of each sequence, each picture, each slice, or each macroblock The moving picture decoding method according to claim 11, wherein the moving picture decoding method is changed to the second inverse transform method. A decoding unit that decodes input encoded data according to a predetermined method and derives orthogonal transform coefficients necessary for decoding processing, prediction mode information, and transform information;
Inverse transform method selection for selecting a first inverse transform method that depends on the prediction direction or a second inverse transform method that does not depend on the prediction direction as an inverse transform method of the decoding target block of the encoded data according to the transform information And
An inverse quantization unit that performs an inverse quantization process on the orthogonal transform coefficient obtained in the decoding step and generates an orthogonal transform coefficient;
According to the inverse transform method selected in the inverse transform method selection step, an inverse orthogonal transform unit that performs inverse orthogonal transform on the orthogonal transform coefficient and generates a decoded prediction error signal;
An image restoration unit that generates a prediction signal from the decoded decoded image signal based on the coding mode information, and adds the prediction signal to the decoded prediction error signal to generate a decoded image;
A moving picture decoding apparatus comprising:

Images