DESCRIPTION IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND
PROGRAM
5 TECHNICAL FIELD [0001]
The present technology relates to an image processing device, an image processing method, and a program, • and more particularly to an image processing 10 device, an image processing method, and a program capable of reducing the amount of computation required for an orthogonal transform process or an inverse orthogonal transform process.
15 BACKGROUND ART
[0002]
An encoding scheme that uses an orthogonal
transform called a rotation transform (ROT) has been
considered as an encoding scheme corresponding to a next-2 0 generation Advanced video coding (AVC) scheme (for
example, see Patent Document 1). The conventional
discrete cosine transform (DCT) which is widely used in
video coding is not optimal in some situations.
For example, when a transform target has a strong 25 directional component, the DCT basis vector may not
satisfactorily express the strong directional component.
[0003]
In general, although directional transform (ROT)
can solve the above problem, it is difficult to perform 30 ROT because the ROT requires many floating point
operations and requires that a transform target block has

SP307551WO00
a square shape. In particular, it is more difficult to apply the ROT when there are a large number of block sizes. [0004] 5 Therefore, a method has been proposed in which a processing unit that performs ROT with a small number of block sizes is designed, and ROT is performed on only a low frequency component as a second transform following the DOT.
10 [0005]
Fig, 1 illustrates an example of the steps of inverse ROT in a decoder that decodes encoded image data by performing orthogonal transform according to such a method.
15 [0006]
White boxes on the left side are image data as residual information that is extracted from encoded image data. The image data is dequantized for respective blocks made up of the pixel values of the 4x4, 8x8, 16x16,
20 32x32, 64x64, or 128x128 pixels. Moreover, only a 4x4 or 8x8 pixel block made up of a low frequency component among the dequantized blocks is subjected to inverse ROT, and the coefficients obtained after the inverse ROT and the remaining high frequency component of the dequantized
25 blocks are subjected to inverse DOT.
[0007] ._- - - - -
By doing so, in the above-described method, it is necessary to prepare only block sizes of 4x4 or 8x8 pixels as a block size for ROT and inverse ROT.
30
CITATION LIST

SP307551WO00
NON-PATENT DOCUMENT [0008]
Non-Patent Document 1: http://wftp3.itu.int/av-arch/jctvc-site/2010_04_A_Dresden/JCTVC-Al24.zip 5 (searched on October 28, 2010)
SUMMARY OF THE INVENTION
PROBLEMS TO BE SOLVED BY THE INVENTION
[0009]
10 However, a problem occurs when the block size of the intra prediction is as small as 4x4 pixels. Specifically, in order to perform intra prediction of the respective blocks, since the decoded image data of the neighboring block including a block to the left of the
15 corresponding block is required, it is difficult to perform intra prediction of the respective blocks in parallel. Moreover, in the above-described method, in order to obtain the decoded image data, a large number of processes such as DOT, ROT, quantization, dequantization,
20 inverse ROT, and inverse DOT are required. [0010]
Thus, when the block size of the intra prediction is small, the longest period required for encoding and decoding macroblocks or coding units (CUs) increases, and
25 it is very difficult to use the above-described method in applications that require real-time properties. Here, the CU is the same concept as the macroblock in the AVC scheme. [0011]
30 The present technology has been made in view of such a circumstance and aims to reduce the amount of

SP307551WO00
processing required for ROT and DCT or inverse DCT and inverse ROT.
SOLUTION TO PROBLEMS 5 [0012]
An image processing device or a program according to an aspect of the present technology is an image processing device including: a dequantization unit that dequantizes a quantized image to obtain a low frequency
10 component having a predetermined size of the image, which is obtained by performing a second orthogonal transform after a first orthogonal transform, and to obtain a high frequency component, which is a component other than the low frequency component of the image and is obtained by
15 the first orthogonal transform; and an inverse orthogonal transform unit that, when a size of the image is the predetermined size, performs a third inverse orthogonal transform, which is a combined transform of a first inverse orthogonal transform corresponding to the first
20 orthogonal transform and a second inverse orthogonal transform corresponding to the second orthogonal transform, on the image which is the low frequency component, and that, when the size of the image is larger than the predetermined size, performs the second inverse
25 orthogonal transform on the low frequency component and performs the first inverse orthogonal transform on the low frequency component having been subjected to the second inverse orthogonal transform and the high frequency component obtained by the dequantization unit,
30 or a program for causing a computer to function as the image processing device.

SP307551WO00
[0013]
An image processing method according to an aspect of the present technology is an image processing method including the steps of: dequantizing a quantized image to 5 obtain a low frequency component having a predetermined size of the image, obtained by performing a second orthogonal transform after a first orthogonal transform and to obtain a high frequency component which is a component other than the low frequency component of the
10 image, obtained by the first orthogonal transform; when the size of the image is the predetermined size, performing a third inverse orthogonal transform, which is a combined transform of a first inverse orthogonal transform corresponding to the first orthogonal transform
15 and a second inverse orthogonal transform corresponding
to the second orthogonal transform, on the image which is the low frequency component; when the size of the image is larger than the predetermined size, performing the second inverse orthogonal transform on the low frequency
20 component; and performing the first inverse orthogonal transform on the low frequency component having been subjected to the second inverse orthogonal transform and the high frequency component obtained by the dequantization unit.
25 [0014]
In the aspect of the present technology, a quantized image is dequantized to obtain a low frequency component having a predetermined size of the image, obtained by performing a second orthogonal transform
30 after a first orthogonal transform and to obtain a high frequency component which is a component other than the

SP307551WO00
low frequency component of the image, obtained by the first orthogonal transform. When the size of the image is the predetermined size, a third inverse orthogonal transform, which is a combined transform of a first 5 inverse orthogonal transform corresponding to the first orthogonal transform and a second inverse orthogonal transform corresponding to the second orthogonal transform, is performed on the image which is the low frequency component. When the size of the image is
10 larger than the predetermined size, the second inverse orthogonal transform is performed on the low frequency component, and the first inverse orthogonal transform is performed on the low frequency component having been subjected to the second inverse orthogonal transform and
15 the high frequency component obtained by the dequantization unit.
EFFECTS OF THE INVENTION [0015] 2 0 According to the aspect of the present technology, it is possible to reduce the amount of processing required for ROT and DCT or inverse DCT and inverse ROT.
BRIEF DESCRIPTION OF DRAWINGS 25 [0016]
Fig. 1 is a diagram illustrating an example, of the steps of ROT in a decoder.
Fig. 2 is a block diagram illustrating a configuration example of an AVC encoder. 30 Fig. 3 is a block diagram illustrating a configuration example of an AVC decoder.

SP307551WO00 >
Fig. 4 is a block diagram illustrating a configuration example of portions corresponding to an orthogonal transformer, a quantizer, a dequantizer, and an inverse orthogonal transformer when ROT is introduced. 5 Fig. 5 is a diagram for describing an improvement of ROT on the encoder side.
Fig. 6 is a block diagram illustrating a configuration example corresponding to portions corresponding to a dequantizer and an inverse orthogonal 10 transformer when ROT is introduced.
Fig. 7 is a diagram for describing an improvement of ROT on the decoder side.
Fig. 8 is a flowchart for describing the processes of the encoder. 15 Fig. 9 is a flowchart for describing the processes of the encoder.
Fig. 10 is a flowchart for describing the processes of the encoder.
Fig. 11 is a flowchart for describing the processes 2 0 of the encoder.
Fig. 12 is a flowchart for describing the processes of the encoder.
Fig. 13 is a flowchart for describing the processes of the decoder. 25 Fig. 14 is a flowchart for describing the processes of the decoder.
Fig. 15 is a flowchart for describing the processes of the decoder.
Fig. 16 is a block diagram illustrating a 30 configuration example of an embodiment of a computer to which the present technology is applied.

SP307551WO00
MODE FOR CARRYING OUT THE INVENTION [0017]
5 [Configuration Example of Encoder]
Fig. 2 is a block diagram illustrating a configuration example of an embodiment of an AVC encoder to which the present technology is applied. [0018]
10 The encoder of Fig. 2 includes an A/D converter 101, a frame rearrangement buffer 102, a computing device 103, an orthogonal transformer 104, a quantizer 105, a lossless encoder 106, a storage buffer 107, a dequantizer 108, an inverse orthogonal transformer 109, an adder 110,
15 a deblocking filter 111, a frame memory 112, a motion compensator 113, an intra predictor 114, a rate controller 115, a motion predictor 116, and a selector 117. The encoder of Fig. 1 compresses and encodes an input image according to the AVC scheme.
20 [0019]
Specifically, the A/D converter 101 of the encoder performs A/D conversion on a frame-based image input as an input signal to obtain digital image data and outputs the digital image data to the frame rearrangement buffer
25 102 which stores the digital image data. The frame
rearrangement buffer 102 rearranges the frames of the image arranged in the stored order for display according to a group of picture (GOP) structure, in the order for encoding.
30 [0020]
The computing device 103 subtracts a prediction

SP307551WO00
image supplied from the selector 117 from the image read from the frame rearrangement buffer 102 as necessary. The computing device 103 outputs an image obtained as a result of the subtraction to the orthogonal transformer 5 104 as residual information. When the prediction image is not supplied from the selector 117, the computing device 103 outputs the image read from the frame rearrangement buffer 102 to the orthogonal transformer 104 as residual information without any change.
10 [0021]
The orthogonal transformer 104 performs an orthogonal transform corresponding to a block size on the residual information from the computing device 103. Specifically, when the block size is 4x4 pixels, the
15 orthogonal transformer 104 performs a combined transform of DCT and ROT on the residual infoirmation. On the other hand, when the block size is 8x8 pixels, the orthogonal transformer 104 performs DCT on the residual infoirmation and performs ROT on coefficients obtained as a result of
20 the DCT. Moreover, when the block size is larger than 8x8 pixels, the orthogonal transformer 104 performs DCT on the residual information, performs ROT on a low frequency component of 8x8 pixels among the coefficients obtained as a result of the DCT, and uses the
25 coefficients obtained as a result of the ROT and a
remaining high frequency, component as final coefficients. The orthogonal transformer 104 supplies coefficients obtained as a result of the orthogonal transform to the quantizer 105.
30 [0022] -
The quantizer 105 quantizes the coefficients

SP307551WO00
supplied from the orthogonal transformer 104. The quantized coefficients are input to the lossless encoder 106. [0023] 5 The lossless encoder 106 acquires information (hereinafter referred to as intra prediction mode information) that indicates an optimal intra prediction mode from the intra predictor 114 and acquires information (hereinafter referred to as inter prediction
10 mode information) that indicates an optimal inter
prediction mode, motion vector information, and the like
from the motion predictor 116.
[0024]
The lossless encoder 106 performs lossless encoding
15 such as variable length coding (for example, Context-Adaptive Variable Length Coding (CAVLC)) or arithmetic coding (for example, Context-Adaptive Binary Arithmetic Coding (CABAC)) on the quantized coefficients supplied from the quantizer 105 to obtain information obtained as
20 a result of the encoding as a compressed image. Moreover, the lossless encoder 106 performs lossless encoding on the intra prediction mode information, the inter prediction mode information, the motion vector information, and the like to obtain information obtained
25 as a result of the encoding as header information that is added to the compressed image. The lossless encoder 10 6 supplies the compressed image to which the header information obtained as a result of the lossless encoding is added to the storage buffer 107 as image compression
30 information. [0025]
10

SP307551WO00
The storage buffer 107 temporarily stores the image compression information supplied from the lossless encoder 106 and outputs the image compression information to a recording device (not illustrated), a transmission 5 path (not illustrated), or the like which is on the downstream side, for example. [0026]
Moreover, the quantized coefficients output from the quantizer 105 are also input to the dequantizer 108 10 and dequantized by the dequantizer 108 and are supplied to the inverse orthogonal transformer 109. [0027]
The inverse orthogonal transformer 109 performs an inverse orthogonal transform corresponding to a block 15 size on the coefficients supplied from the dequantizer 108. Specifically, when the block size is 4x4 pixels, the inverse orthogonal transformer 109 performs a combined transform of inverse ROT and inverse DOT on the coefficients. On the other hand, when the block size is
20 8x8 pixels, the inverse orthogonal transformer 109
performs inverse ROT on the coefficients and performs inverse DCT on the coefficients obtained as a result of the inverse ROT. Moreover, when the block size is larger than 8x8 pixels, the inverse orthogonal transformer 109
25 performs inverse ROT on an 8x8 low frequency component of
the coefficients and performs-inverse DOT^on^the—
coefficients obtained as a result of the inverse ROT and the remaining high frequency component. The inverse orthogonal transformer 109 supplies the residual
30 information obtained as a result of the inverse orthogonal transform to the adder 110.
11

SP307551WO00
[0028]
The adder 110 adds the residual information supplied from the inverse orthogonal transformer 109 to the prediction image supplied from the intra predictor 5 114 or the motion compensator 113 as necessary and
obtains a locally decoded image. The adder 110 supplies the obtained image to the deblocking filter 111 and supplies the obtained image to the intra predictor 114 as a reference image.
10 [0029]
The deblocking filter 111 performs filtering on the locally decoded image supplied from the adder 110 to thereby remove a block distortion. The deblocking filter 111 supplies the image obtained as a result of the
15 filtering to the frame memory 112, which stores the image. The image stored in the frame memory 112 is output to the motion compensator 113 and the motion predictor 116 as a reference image. [0030]
20 The motion compensator 113 performs a compensation process on the reference image supplied from the frame memory 112 based on the motion vector and the inter prediction mode information supplied from the motion predictor 116 to generate a prediction image. The motion
25 compensator 113 supplies a cost function value (details of which will be described—later) supplied from the— motion predictor 116 and the generated prediction image to the selector 117. [0031]
30 The cost function value is also referred to as a Rate Distortion (RD) cost, and is calculated based on a
12

SP307551WO00
High Complexity mode or a Low Complexity mode as defined in Joint Model (JM) which is reference software of the AVC scheme, for example. [0032] 5 Specifically, when the High Complexity mode is used as a method of calculating the cost function value, processes up to lossless encoding are temporarily performed on all candidate prediction modes, and a cost function value expressed by Expression (1) below is 10 calculated for each prediction mode. [0033]
Cost (Mode) = D + X-R ... (1) [0034]
Here, "D" is a difference (distortion) between an 15 original image and a decoded image, "R" is an occurrence coding rate including the orthogonal transform
coefficients, and "A," is the Lagrange's multiplier given
as a function of a quantization parameter QP.
[0035]
20 On the other hand, when the Low Complexity mode is used as a method of calculating the cost function value, generation of a decoded image and calculation of a header bit such as information that indicates a prediction mode are performed on all candidate prediction modes, and a
25 cost function expressed by Expression (2) below is calculated for each prediction mode. [0036]
Cost(Mode) = D + QPtoQuant (QP) •Header_Bit ... (2) [0037]
30 Here, "D" is a difference (distortion) between an original image and a decoded image, "Header Bit" is
13

SP307551WO00
i
header bit of a prediction mode, and "QPtoQuant" is a function given as a function of a quantization parameter
QP.
[0038] 5 In the Low Complexity mode, since it is only
necessary to generate a decoded image in all prediction modes, and it is not necessary to perform lossless encoding, a small amount of computation is required. In this example, it is assumed that the High Complexity mode
10 is used as the method of calculating the cost function value. [0039]
The intra predictor 114 performs an intra prediction process in all candidate intra prediction
15 modes in units of blocks of all candidate block sizes based on the image read from the frame rearrangement buffer 102 and the reference image supplied from the adder 110 to generate a prediction image. [0040]
20 Moreover, the intra predictor 114 calculates the cost function value for all candidate intra prediction modes and all candidate block sizes. Moreover, the intra predictor 114 determines a combination of an intra prediction mode and a block size in which the cost
25 function value is smallest as an optimal intra prediction mode. The intra predictor 114 supplies- the prediction image generated in the optimal intra prediction mode and the corresponding cost function value to the selector 117 When selection of a prediction image generated in the
30 optimal intra prediction mode is notified from the
selector 117, the intra predictor 114 supplies the intra
14

SP307551WO00
prediction mode information to the lossless encoder 106. [0041]
The motion predictor 116 performs motion prediction in all candidate inter prediction modes based on the 5 image supplied from the frame rearrangement buffer 102 and the reference image supplied from the frame memory 112 to generate a motion vector. In this case, the motion predictor 116 calculates a cost function value in all candidate inter prediction modes and determines an
10 inter prediction mode in which the cost function value is smallest as an optimal inter prediction mode. Moreover, the motion predictor 116 supplies the inter prediction mode information and the corresponding motion vector and cost function value to the motion compensator 113. When
15 selection of a prediction image generated in the optimal inter prediction mode is notified from the selector 117, the motion predictor 116 outputs the inter prediction mode information, information on the corresponding motion vector, and the like to the lossless encoder 106.
20 [0042]
The selector 117 determines any one of the optimal intra prediction mode and the optimal inter prediction mode as an optimal prediction mode based on the cost function value supplied from the intra predictor 114 and
25 the motion compensator 113. Moreover, the selector 117 supplies the prediction'image in the optimal prediction mode to the computing device 103 and the adder 110. Moreover, the selector 117 notifies selection of the prediction image in the optimal prediction mode to the
30 intra predictor 114 or the motion predictor 116. [0043]
15

SP307551WO00
The rate controller 115 controls the rate of the quantization operation of the quantizer 105 based on the image compression information stored in the storage buffer 107 so that an overflow or an underflow does not 5 occur. [0044] [Configuration Example of Decoder]
Fig. 3 is a block diagram of an AVC decoder corresponding to the encoder of Fig. 2.
10 [0045]
The decoder of Fig. 3 includes a storage buffer 216, a lossless decoder 217, a dequantizer 218, an inverse orthogonal transformer 219, an adder 220, a frame rearrangement buffer 221, a D/A converter 222, a frame
15 memory 223, a motion compensator 224, an intra predictor 225, a deblocking filter 226, and a switch 227. [0046]
The storage buffer 216 stores the image compression information transmitted from the encoder of Fig. 2. The
20 lossless decoder 217 reads and acquires the image
compression information from the storage buffer 216 and losslessly decodes the image compression information according to a scheme corresponding to the lossless encoding scheme of the lossless encoder 106 of Fig. 2.
25 [0047]
Specifically, the lossless decoder 217 losslessly decodes the header information in the image compression information to acquire the intra prediction mode information, the inter prediction mode information, the
30 motion vector information, and the like. Moreover, the lossless decoder 217 losslessly decodes the compressed
16

SP307551WO00
image in the image compression information. [0048]
Moreover, the lossless decoder 217 supplies quantized coefficients obtained as a result of lossless 5 decoding of the compressed image to the dequantizer 218. The lossless decoder 217 supplies the intra prediction mode information obtained as a result of the lossless decoding to the intra predictor 225 and supplies the inter prediction mode information, the motion vector
10 information, and the like to the motion compensator 224. [0049]
The dequantizer 218 has the same configuration as the dequantizer 108 of Fig. 2 and dequantizes the quantized coefficients supplied from the lossless decoder
15 217 according to a scheme corresponding to the
quantization scheme of the quantizer 105 of Fig. 2. The dequantizer 218 supplies the coefficients obtained as a result of the dequantization to the inverse orthogonal transformer 219.
20 [0050]
The inverse orthogonal transformer 219 performs an inverse orthogonal transform corresponding to a block size on the coefficients supplied from the dequantizer 218 in a manner similarly to the inverse orthogonal
25 transformer 109 of Fig. 2. The inverse orthogonal transformer 219 .supplies the residual information obtained as a result of the inverse orthogonal transform to the adder 220. [0051]
30 The adder 220 adds the residual information
supplied from the inverse orthogonal transformer 219 to
17

SP307551WO00
the prediction image supplied from the switch 227 and decodes the added result as necessary. The adder 220 supplies a decoded image obtained as a result of the decoding to the intra predictor 225 and the deblocking 5 filter 226. [0052]
The deblocking filter 22 6 performs filtering on the decoded image supplied from the adder 220 to thereby remove a block distortion. The deblocking filter 22 6
10 supplies an image obtained as a result of the filtering to the frame memory 223, which stores the image, and outputs the image to the frame rearrangement buffer 221. [0053]
The frame rearrangement buffer 221 rearranges the
15 image supplied from the deblocking filter 22 6.
Specifically, the order of frames of the image arranged for encoding by the frame rearrangement buffer 102 of Fig. 2 is rearranged to the original display order. The D/A converter 222 performs D/A conversion on the image
20 rearranged by the frame rearrangement buffer 221 and outputs the converted image to a display (not illustrated), which displays the image. [0054]
The frame memory 223 reads the image stored therein
25 as a reference image and outputs the reference image to the motion compensator 224. -[0055]
The intra predictor 225 performs an intra prediction process in the optimal intra prediction mode
30 indicated by the intra prediction mode information based on the intra prediction mode information supplied from
18

SP307551WO00 I
the lossless decoder 217 to generate a prediction image.
The intra predictor 225 supplies the prediction image to
the switch 227.
[0056] 5 The motion compensator 224 performs a motion
compensation process on the reference image supplied from
the frame memory 223 based on the inter prediction mode
information, the motion vector information, and the like
supplied from the lossless decoder 217 to generate a 10 prediction image. The motion compensator 224 supplies
the prediction image tO' the switch 227.
[0057]
The switch 227 selects the prediction image
generated by the motion compensator 224 or the intra 15 predictor 225 and supplies the s'elected prediction image
to the adder 220.
[0058]
Description of Orthogonal Transform and Inverse
Orthogonal Transform 20 First, Fig. 4 is a block diagram illustrating an
orthogonal transformer, a quantizer, a dequantizer, and
an inverse orthogonal transformer of a conventional
encoder when DCT and ROT are performed as an orthogonal
transform. 25 [0059]
As illustrated in Fig. 4, an orthogonal transformer
of the conventional encoder includes a 4x4 DCT 411, an
8x8 DCT 412, a 16x16 DCT 413, a 32x32 DCT 414, a 64x64
DCT 415, a 128x128 DCT 416, a 4x4 ROT 417, and an 8x8 ROT 30 418.
[0060]
19

SP307551WO00
The residual information is input to the 4x4 DCT 411, the 8x8 DCT 412, the 16x16 DCT 413, the 32x32 DCT 414, the 64x64 DCT 415, and the 128x128 DCT 416 according to the block size, and is subjected to DCT. 5 [0061]
Specifically, the 4x4 DCT 411 performs DCT on the 4x4 pixel residual information, rounds off the computation accuracy of the 4x4 pixel coefficients obtained as a result of the DCT, and supplies the 4x4.
10 pixel coefficients to the 4x4 ROT 417. [0062]
The 8x8 DCT 412 performs DCT on the 8x8 pixel residual information, rounds off the computation accuracy of the 8x8 pixel coefficients obtained as a result of the
15 DCT, and supplies the 8x8 pixel coefficients to the 8x8 ROT 418. The 16x16 DCT 413 performs DCT on the 16x16 pixel residual information and rounds off the computation accuracy of the 16x16 pixel coefficients obtained as a result of the DCT. The 16x16 DCT 413 supplies an 8x8
20 pixel low frequency component among the 16x16 pixel
coefficients obtained as a result of the DCT to the 8x8 ROT 418 and supplies the remaining high frequency component to the quantizer. [0063]
25 Similarly, the 32x32 DCT 414, the 64x64 DCT 415,
and the 128x128 DCT 416 perform DCT on the 32x32, 64x64, and 128x128 pixel residual information, respectively, and round off the computation accuracy of the coefficients obtained as a result of the DCT. Moreover, the 32x32 DCT
30 414, the 64x64 DCT 415, and the 128x128 DCT 416 supplies only the 8x8 pixel low frequency component among the
20

SP307551WO00

coefficients obtained as a result of the DCT to the 8x8 ROT 418 and supplies the remaining high frequency-component to the quantizer. [0064] 5 The 4x4 ROT 417 performs ROT on the 4x4 pixel coefficients supplied from the 4x4 DCT 411 using an angular index. [0065]
Here, the ROT is a rotational transform that uses a 10 rotation matrix Rverticai for vertical direction arid a rotation matrix Rhorizontai for horizontal direction illustrated in Expression (1) below, and the angular index is ai to as in Expression (1) . [0066] 15 [Mathematical Formula 1]
cosQficosQ;3-sinaicosQ[2sinff3 -smQ;iCosff3-cosQriCosci!2siBa3 siiiQf2sinQ!3 0
Rvolicjl(ff|>ff2>fl!3) =
cosffisifflffs+siiiaiCosQfiCosffa -sinffisinofs+cosaicosaicosffs -sinOiCosffs 0
sinffisina2 cosCfisinClfi cosQfi 0
0 0 0

Rhori2)rtil(ff|» 05)06) ■

cosff4cosffrsinff-(cosff5smff6 -smflf^sfl6-cosQf4COSQ!5sinff6 sinffssinffj 0
cosfl4sinff«+siiiQf4COSff5Cosflf6 -sinQf^smffe+cosQf^cosQfsCOSffe -slnflfscosofj 0
sinQf4sma5 cosff4sinQf5 cosQfs 0
0 0 0 1

20

•••(1)
[0067]
The 8x8 ROT 418 performs ROT using an angular index on the 8x8 pixel coefficients supplied from the 8x8 DCT 412, the 16x16 DCT 413, the 32x32 DCT 414, the 64x64 DCT 415, and the 128x128 DCT 416. [0068]
The 4x4 pixel coefficients obtained as a result of

21

SP307551WO00
the ROT by the 4x4 ROT 417 and the 8x8 pixel coefficients obtained as a result of the ROT by the 8x8 ROT 418 are supplied to the quantizer with the computation accuracy rounded off. 5 [0069]
The quantizer includes a 4x4 Quant 419, an 8x8 Quant 420, a 16x16 Quant 421, a 32x32 Quant 422, a 64x64 Quant 423, and a 128x128 Quant 424. [0070]
10 The 4x4 Quant 419 quantizes the 4x4 pixel
coefficients supplied from the 4x4 ROT 417. The 4x4 Quant 419 supplies the quantized 4x4 pixel coefficients to the dequantizer and supplies the same to the same lossless encoder (not illustrated) as the lossless
15 encoder 106 of Fig. 2. [0071]
The 8x8 Quant 420 quantizes the 8x8 pixel coefficients supplied from the 8x8 ROT 418. The 8x8 Quant 420 supplies the quantized 8x8 pixel coefficients
20 to the dequantizer and supplies the same to the same lossless encoder (not illustrated) as the lossless encoder 106 of Fig. 2. [0072]
The 16x16 Quant 421 quantizes the 8x8 pixel
25 coefficients supplied from the 8x8 ROT 418 and a high frequency component other than the 8x8 pixel.low frequency component among the coefficients obtained as a result of the DCT on the 16x16 pixel residual information supplied from the 16x16 DCT 413. The 16x16 Quant 421
30 supplies the quantized 16x16 pixel coefficients to the dequantizer and supplies the same to the same lossless
22

SP307551WO00
encoder (not illustrated) as the lossless encoder 105. [0073]
Similarly, the 32x32 Quant 422, the 64x64 Quant 423, and the 128x128 Quant 424 quantize the 8x8 pixel 5 coefficients supplied from the 8x8 ROT 418 and a high frequency component other than the 8x8 pixel low frequency component among the coefficients obtained as a result of the DCT on the 32x32, 64x64, and 128x128 pixel residual information. The 32x32 Quant 422, the 64x64
10 Quant 423, and the 128x128 Quant 424 supply the quantized 32x32, 64x64, and 128x128 pixel coefficients to the dequantizer and supply the same to the same lossless encoder (not illustrated) as the lossless encoder 106. [0074]
15 The dequantizer includes a 4x4 Inv Quant 451, an 8x8 Inv Quant 452, a 16x16 Inv Quant 453, a 32x32 Inv Quant 454, a 64x64 Inv Quant 455, and a 128x128 Inv Quant 456. [0075]
20 The 4x4 Inv Quant 451, the 8x8 Inv Quant 452, the 16x16 Inv Quant 453, the 32x32 Inv Quant 454, the 64x64 Inv Quant 455, and the 128x128 Inv Quant 456 dequantize the quantized coefficients supplied from the 4x4 Quant 419, the 8x8 Quant 420, the 16x16 Quant 421, the 32x32
25 Quant 422, the 64x64 Quant 423, and the 128x128 Quant 424 respectively, and supply the dequantized coefficients to. the inverse orthogonal transformer. [0076]
The inverse orthogonal transformer includes a 4x4
30 Inv ROT 457, an 8x8 Inv ROT 458, a 4x4 Inv DCT 459, an
8x8 Inv DCT 460, a 16x16 Inv DCT 461, a 32x32 Inv DCT 462,
23

SP307551WO00
a 64x64 Inv DCT 463, and a 128x128 Inv DCT 464. [0077]
The 4x4 Inv ROT 457 performs inverse ROT on the dequantized 4x4 pixel coefficients supplied from the 4x4 5 Inv Quant 451 using an angular index. The 4x4 Inv ROT 457 supplies the 4x4 pixel coefficients obtained as a result of the inverse ROT to the 4x4 Inv DCT 459. [0078]
The 8x8 Inv ROT 458 performs inverse ROT on the
10 dequantized 8x8 pixel coefficients supplied from the 8x8 Inv Quant 452 using an angular index and supplies the 8x8 pixel coefficients obtained as a result of the inverse ROT to the 8x8 Inv DCT 4 60. [0079]
15 Moreover, the 8x8 Inv ROT 458 performs inverse ROT on the 8x8 pixel low frequency component among the dequantized 16x16 pixel coefficients supplied from the 16x16 Inv Quant 453 using an angular index. Moreover, the 8x8 Inv ROT 458 supplies the 8x8 pixel coefficients
20 obtained as a result of the inverse ROT to the 16x16 Inv DCT 461. [0080]
Similarly, the 8x8 Inv ROT 458 performs inverse ROT on the 8x8 pixel low frequency component among the
25 dequantized 32x32, 64x64, and 128x128 pixel coefficients supplied from the 32x32- Inv Quant 454, the 64x64 Inv Quant 455, and the 128x128 Inv Quant 456, respectively, using an angular index. Moreover, the 8x8 Inv ROT 458 supplies the 8x8 pixel coefficients obtained as a result
30 of the inverse ROT on the 8x8 pixel low frequency
component among the dequantized 32x32, 64x64, and 128x128
24

SP307551WO00
pixel coefficients to the 32x32 Inv DCT 4 62, the 64x64 Inv DCT 463, and the 128x12 8 Inv DCT 4 64, respectively. [0081]
The 4x4 Inv DCT 459 performs inverse DCT on the 4x4 5 pixel coefficients supplied from the 4x4 Inv Rot 457. The 4x4 Inv DCT 459 supplies the 4x4 pixel residual information obtained as a result of the inverse DCT to the same adder (not illustrated) as the adder 110 of Fig. 2.
10 [0082]
The 8x8 Inv DCT 460 performs inverse DCT on the 8x8 pixel coefficients supplied from the 8x8 Inv Rot 458. The 8x8 Inv DCT 460 supplies the 8x8 pixel residual information obtained as a result of the inverse DCT to
15 the same adder (not illustrated) as the adder 110. The 16x16 Inv DCT 461 performs inverse DCT on the 8x8 pixel coefficients supplied from the 8x8 Inv Rot 458 and a high frequency component other than the 8x8 pixel low frequency component among the 16x16 pixel coefficients
20 supplied from the 16x16 Inv Quant 453. The 16x16 Inv DCT 4 61 supplies the 16x16 pixel residual information obtained as a result of the inverse DCT to the same adder (not illustrated) as the adder 110. [0083]
25 Similarly, the 32x32 Inv DCT 462, the 64x64 Inv DCT 463, and the 128x128:Inv DCT 464 perform inverse.DCT on the 8x8 pixel coefficients supplied from the 8x8 Inv Rot 458 and the high frequency component other than the 8x8 pixel low frequency component among the coefficients
30 supplied from the 32x32 Inv Quant 454, the 64x64 Inv
Quant 455, and the 128x128 Inv Quant 456. The 32x32 Inv
25

ST»307551WOOO
DCT 462, the 64x64 Inv DCT 463, and the 128x128 Inv DCT 464 supply the 32x32, 64x64, and 128x128 pixel residual information obtained as a result of the inverse DCT to the same adder (not illustrated) as the adder 110. 5 [0084]
In this way, the residual information is input to the adder (not illustrated), whereby a decoded image is obtained. [0085]
10 Next, Fig. 5 is a block diagram illustrating the
details of the orthogonal transformer 104, the quantizer 105, the dequantizer 108, and the inverse orthogonal transformer 109 of the encoder of Fig. 2. [0086]
15 Among the configurations illustrated in Fig. 5, the same configurations as the configurations of Fig. 4 are denoted by the same reference numerals. Redundant description thereof will be appropriately not provided. [0087]
2 0 The configuration of Fig. 5 is mainly different
from the configuration of Fig. 4, in that a 4x4 DCTxROT 501 is provided instead of the 4x4 DCT 411 and the 4x4 ROT 417 of the orthogonal transformer 104, and that a 4x4 Inv ROTxInv DCT 502 is provided instead of the 4x4 Inv
25 ROT 457 and the 4x4 Inv DCT 459 of the inverse orthogonal
transformer 109. :...-- ; - ■ -
[0088]
The 4x4 DCTxROT ,501 of the orthogonal tra»sformer 1Q4 performs a combined transform of DCT and ROT on the
30 4x4 pixel residual information supplied from the
computing device 103 of Fig. 2 using an angular index.
26

SP307551WO00
Specifically, the 4x4 DCTxROT 501 is provided with a matrix for a combined transform of DCT and ROT corresponding to an angular index, and the 4x4 DCTxROT 501 obtains 4x4 pixel coefficients after DCT and ROT 5 through one transform using the matrix. The 4x4 DCTxROT 501 supplies the 4x4 pixel coefficients to the 4x4 Quant 419 with the computation accuracy rounded off. [0089]
The DCT and ROT are one kind of orthogonal
10 transform and are generally performed by a matrix
operation. Thus, a matrix for a combined transform of DCT and ROT is a matrix obtained by the product of the matrix used in the matrix operation of DCT and the matrix used in the matrix operation of ROT.
15 [0090]
As described above, in the orthogonal transformer 104, since DCT and ROT can be performed through one transform on the 4x4 pixel residual information, it is possible to reduce the amount of computation required for
20 the orthogonal transform as compared to the orthogonal transformer of Fig. 4. Moreover, since the rounding of the computation accuracy after DCT is not necessary, it is possible to increase the computation accuracy as compared to the orthogonal transformer of Fig. 4. Thus,
25 the output of the 4x4 ROT 417 of Fig. 4 is not the same as the output of the-4x4--DCTxROT 501 of Fig. 5. [0091]
Moreover, the 4x4 Inv ROTxInv DCT 502 of the inverse orthogonal transformer 109 performs a combined
30 transform of inverse DCT and inverse ROT on the 4x4 pixel coefficients supplied from the 4x4 Inv Quant 451 using an
27

SP307551WO00
angular index. Specifically, the 4x4 Inv ROTxInv DCT 502 is provided with a matrix for a combined transform of inverse DCT and inverse ROT corresponding to an angular index, and the 4x4 Inv ROTxInv DCT 502 obtains 4x4 pixel 5 residual information after inverse DCT and inverse ROT through one transform using the matrix. The combined transform of inverse DCT and inverse ROT is an inverse transform of the transform performed by the 4x4 DCTxROT 501. The 4x4 Inv ROTxInv DCT 502 supplies the 4x4 pixel
10 residual information obtained as a result of the transform to the adder 110 of Fig. 2. [0092]
As described above, in the inverse orthogonal transformer 109, since inverse DCT and inverse ROT can be
15 performed through one transform on the 4x4 pixel
coefficients, it is possible to reduce the amount of computation required for the inverse orthogonal transform as compared to the inverse orthogonal transformer of Fig. 4. Moreover, since the rounding of the computation
2 0 accuracy after inverse ROT is not necessary, it is
possible to increase the computation accuracy as compared to the inverse orthogonal transformer of Fig. 4. Thus, the output of the 4x4 Inv DCT 459 of Fig. 4 is not the same as the output of the 4x4 Inv ROTxInv DCT 502 of Fig.
25 5.
[0093] ,-__ . ■
Next, Fig. 6 is a block diagram illustrating a dequantizer and an inverse orthogonal transformer of the conventional decoder when DCT and ROT are performed as an
30 orthogonal transform. [0094]
28

SP307551WO00
The dequantizer of the conventional decoder of Fig. 6 has the same configuration as the dequantizer of Fig. 4, and the inverse orthogonal transformer of Fig. 6 has the same configuration as the inverse orthogonal transformer 5 of Fig. 4. [0095]
Specifically, the dequantizer of Fig. 6 includes a 4x4 Inv Quant 601, an 8x8 Inv Quant 602, a 16x16 Inv Quant 603, a 32x32 Inv Quant 604, a 64x64 Inv Quant 605,
10 and a 128x128 Inv Quant 606. The 4x4 Inv Quant 601, the 8x8 Inv Quant 602, the 16x16 Inv Quant 603, the 32x32 Inv Quant 604, the 64x64 Inv Quant 605, and the 128x128 Inv Quant 606 perform dequantization on the quantized coefficients obtained as a result of lossless decoding of
15 the losslessly encoded image compression information
transmitted from the encoder in a manner similarly to the
dequantizer of Fig. 4
[0096]
Moreover, the inverse orthogonal transformer of Fig.
20 6 includes a 4x4 Inv ROT 607, an 8x8 Inv ROT 608, a 4x4 Inv DOT 609, an 8x8 Inv DCT 610, a 16x16 Inv DCT 611, a 32x32 Inv DCT 612, a 64x64 Inv DCT 613, and a 128x128 Inv DCT 614. The 4x4 Inv ROT 607 and the 8x8 Inv ROT 608 perform inverse ROT in a manner similarly to the 4x4 Inv
25 ROT 457 and the 8x8 Inv ROT 458 of Fig. 4, respectively. Moreover, the 4x4 Inv DCT 609, the 8x8 Inv DCT 610, the 16x16 Inv DCT 611, the 32x32 Inv DCT 612, the 64x64 Inv DCT 613, and the 12-8x128 Inv DCT 614 perform inverse DCT in a manner similarly to the Inv DCTs of the
30 corresponding block sizes of Fig. 4, respectively. [0097]
29

SP307551WO00
Next, Fig. 7 is a block diagram illustrating the details of the dequantizer 218 and the inverse orthogonal transformer 219 of the decoder of Fig. 3. [0098] 5 The dequantizer 218 of Fig. 7 has the same
configuration as the dequantizer 108 of Fig. 5, and the inverse orthogonal transformer 219 of Fig. 7 has the same configuration as the inverse orthogonal transformer 109 of Fig. 5.
10 [0099]
Among the configurations illustrated in Fig. 7 the same configurations as the configurations of Fig. 6 are denoted by the same reference numerals. Redundant description thereof will be appropriately not provided.
15 [0100]
The configuration of Fig. 7 is mainly different from the configuration of Fig. 6, in that a 4x4 Inv ROTxInv DCT 701 is provided instead of the 4x4 Inv ROT 607 and the 4x4 Inv DCT 609 of the inverse orthogonal
20 transformer 219 similarly to the inverse orthogonal transformer 109. [0101]
The 4x4 Inv ROTxInv DCT 701 of the inverse orthogonal transformer 219 performs a combined transform
25 of inverse OCT and inverse ROT on the 4x4 pixel
coefficients supplied from" the" 4x4 Inv Quant 601 using an angular index in a manner- similarly to the 4x4 Inv ROTxInv DCT 502 of Fig. 5. The 4x4 Inv ROTxInv DCT 701 supplies the 4x4 pixel residual information obtained as a
30 result of the transform to the adder 220 of Fig. 3. [0102]
30

SP307551WO00
The angular index is determined by the encoder, for example, and is included in the header information by the lossless encoder 106 and transmitted to the decoder. [0103] 5 In the present embodiment, although the DCT and ROT on the 4x4 pixel residual information are performed through one transform, the DCT and ROT on the 8x8 pixel residual information as well as the 4x4 pixel residual information may be performed through one transform. The
10 same is true for the inverse DCT and inverse ROT. [0104]
Moreover, in the present embodiment, although the ROT is performed on only the low frequency component of the 8x8 pixels among the coefficients having a size of
15 8x8 pixels or larger obtained as a result of the DCT, the maximum size of the coefficients subjected to the ROT may be different from the size of 8x8 pixels (for example, 4x4 pixels, 16x16 pixels, or the like). The same is true for the inverse ROT.
20 [0105]
Description of Encoder Processing
Figs. 8, 9, 10, 11, and 12 are flowcharts of the processing of the encoder of Fig. 2. [0106]
25 Fig. 8 is a flowchart for describing a macroblock (MB) encoding process. -•----. [0107]
In step Sll of Fig. 8, the encoder calculates a RD cost (P) when inter prediction is used. The details of
30 the process of calculating a RD cost (P) when inter
prediction is used will be described with reference to
31

SP307551WO00
Fig. 9 described later. [0108]
In step S12, the encoder calculates a RD cost (I) when an intra prediction is used. The details of the 5 process of calculating a RD cost (I) when intra
prediction is used will be described with reference to
Fig. 12 described later.
[0109]
In step S13, the selector 117 determines whether
10 the RD cost (I) is larger than the RD cost (P). [0110]
When it is determined in step S13 that the RD cost (I) is not larger than the RD cost (P), that is, when the RD cost (I) is equal to or smaller than the RD cost (P),
15 the selector 117 determines an optimal intra prediction mode as an optimal prediction mode. Moreover, the selector 117 supplies the prediction image in the optimal intra prediction mode to the computing device 103 and the adder 110. Moreover, the selector 117 notifies the
20 selection of the prediction image in the optimal intra
prediction mode to the intra predictor 114. In this way, the intra predictor 114 supplies the intra prediction mode information to the lossless encoder 106. [0111]
25 In step S14, the encoder encodes a current
macroblock (the MB) according to. intra prediction in the optimal intra prediction mode. Specifically, the computing device 103 of the encoder subtracts the prediction image supplied from the selector 117 from the
30 current macroblock of the image read from the frame
rearrangement buffer 102, and the orthogonal transformer
32

SP307551WO00
104 performs orthogonal transform on the residual information obtained as a result of the subtraction. The quantizer 105 quantizes the coefficients obtained as a result of the orthogonal transform of the orthogonal 5 transformer 104, and the lossless encoder 106 losslessly encodes the quantized coefficients and losslessly encodes the intra prediction mode information or the like to be used as the header information. The storage buffer 107 temporarily stores the compressed image, in which the
10 header information obtained as a result of the lossless encoding is added, as the image compression information, and outputs the image compression information. [0112]
On the other hand, when it is determined in step
15 S13 that the RD cost (I) is larger than the RD cost (P) , the selector 117 determines the optimal inter prediction mode as an optimal prediction mode. Moreover, the selector 117 supplies the prediction image in the optimal inter prediction mode to the computing device 103 and the
20 adder 110. Moreover, the selector 117 notifies the
selection of the prediction image in the optimal inter prediction mode to the motion predictor 116. In this way, the motion predictor 116 outputs the inter prediction mode information, the corresponding motion vector
25 information, and the like to the lossless encoder 106. [0113]
In step S15, the encoder encodes a current macroblock according to inter prediction in the optimal inter prediction mode. Specifically, the computing
30 device 103 of. the encoder subtracts the prediction image supplied from the selector 117 from the current
33

SP307551WO00
macroblock of the image read from the frame rearrangement buffer 102, and the orthogonal transformer 104 performs orthogonal transform on the residual information obtained as a result of the subtraction. The quantizer 105 5 quantizes the coefficients obtained as a result of the orthogonal transform of the orthogonal transformer 104, and the lossless encoder 106 losslessly encodes the quantized coefficients and losslessly encodes the inter prediction mode information, the motion vector
10 information, and the like to be used as the header
information. The storage buffer 107 temporarily stores the compressed image in which the header information obtained as a result of the lossless encoding as the image compression information and outputs the image
15 compression information. [0114]
Fig. 9 is a flowchart for describing the details of the process of calculating a RD cost (P) when the inter prediction of step Sll of Fig. 8 is used.
20 [0115]
In step S31 of Fig. 9, the motion predictor 116 sets the block size of the inter prediction to one which has not been set among the 4x4, 8x8, 16x16, 32x32, 64x64, and 12 8x128 pixels corresponding to the respective inter
25 prediction modes. [0116]
In step S32, the motion predictor 116 performs motion prediction with the size set in step S31. Specifically, the motion predictor 116 performs motion
30 prediction in respective blocks of the size set in step
S31 using the image supplied from the frame rearrangement
34

SP307551WO00

buffer 102 and the reference image supplied from the frame memory 112. As a result, motion vectors (MV) for respective blocks are obtained. The motion predictor 116 supplies the motion vector to the motion compensator 113. 5 [0117]
In step S33, the motion compensator 113 performs motion compensation (MC) according to the motion vector supplied from the motion predictor 116. Specifically, the motion compensator 113 generates a prediction image
10 from the reference image supplied from the frame memory 112 according to the motion vector. The motion compensator 113 supplies the generated prediction image to the computing device 103 via the selector 117. [0118]
15 In step S34, the" computing device 103 computes a difference between the image corresponding to the input signal and the MC image (prediction image). The computing device 103 supplies the difference obtained as a result of the computation to the orthogonal transformer
20 104 as residual information. [0119]
In step S35, the orthogonal transformer 104 sets the angular index to one which has not been set among the angular indices of index numbers 0, 1, 2, and 3. The
25 index niimber is a number unique to the combination of the
angular indices tti to a^, and in the present embodiment, a combination of four angular indices of the nimnbers 0 to 3 is prepared. [0120] 30 In step S36, the orthogonal transformer 104
performs a ROT process or the like which is a process of
35

SP307551WO00
performing ROT according to an angular index with respect to the residual information (difference information) supplied from the computing device 103. The details of the process of step S36 will be described with reference 5 to Fig. 10 described later. [0121]
In step S37, the quantizer 105 performs a quantization process which is a process of quantizing the coefficients obtained as a result of the ROT process or
10 the like in step S36. Specifically, the 4x4 Quant 419, the 8x8 Quant 420, the 16x16 Quant 421, the 32x32 Quant 422, the 64x64 Quant 423, or the 128x128 Quant 424 corresponding to the block size of the. inter prediction of the quantizer 105 quantizes the coefficients supplied
15 from the orthogonal transformer 104. The quantizer 105 supplies the coefficients obtained as a result of the quantization process to the lossless encoder 106 and the dequantizer 108. [0122]
20 In step S38, the lossless encoder 106 losslessly encodes the coefficients (quantized coefficients) supplied from the quantizer 105 to obtain a compressed image. [0123]
25 In step S39, the dequantizer 108 performs a
dequantization process which is a process of dequantizing the coefficients supplied from the quantizer 105. Specifically, the 4x4 Inv Quant 451, the 8x8 Inv Quant 452, the 16x16 Inv Quant 453, the 32x32 Inv Quant 454,
30 the 64x64 Inv Quant 455, or the 128x128 Inv Quant 456
corresponding to the block size of the inter prediction
36

SP307551WO00
of the dequantizer 108 dequantizes the coefficients supplied from the quantizer 105. The coefficients obtained as a result of the dequantization process are supplied to the inverse orthogonal transformer 109. 5 [0124]
In step S40, the inverse orthogonal transformer 109 performs an inverse ROT process or the like which is a process of performing inverse ROT according to the angular index set in step S35 with respect to the
10 coefficients corresponding to the residual information
(difference information). The details of the process of step S40 will be described with reference to Fig. 11 described later. [0125]
15 After the process of step S40 is performed, the
flow returns to step S35, and the processes of steps S35 to S40 are repeatedly performed until all of the angular indices of the index numbers 0 to 3 are set as the angular index. Moreover, when all of the angular indices
20 of the index numbers 0 to 3 are set as the angular index, the flow returns to step S31, Moreover, the processes of steps S31 to S40 are repeatedly performed until all sizes
of the 4x4, 8x8, 16x16, 32x32, 64x64, and 128x128 pixels are set as the block size of the inter prediction.
25 [0126]
Moreover, when all sizes of the 4x4, 8x8, 16x16, 32x32, 64x64, and 128x128 pixels are set as the block size of the inter prediction, and all of the angular indices of the index numbers 0 to 3 are set as the
30 angular index with respect to the inter prediction block of each block size, the flow proceeds to step S41.
37

•

SP307551WO00
[0127]
In step S41, the motion predictor 116 computes a RD cost from the MV information, the quantized code information, the decoded image with respect to each 5 combination of the inter prediction mode and the angular index. Specifically, the motion predictor 116 generates a prediction image using the motion vector and the reference image supplied from the frame memory 112 with respect to each combination of the inter prediction mode
10 and the angular index. Moreover, the motion predictor 116 computes a difference between the prediction image and the image supplied from the frame rearrangement buffer 102. Moreover, the motion predictor 116 computes Expression (1) described above and calculates the RD cost
15 using the difference, the occurrence coding amount of the compressed image obtained by the process of step S38, and the like. [0128]
Moreover, the motion predictor 116 uses the
20 smallest RD cost among the RD costs of the respective
combinations of the inter prediction mode corresponding to the block size of the inter prediction and the angular index as the RD cost (P). That is, the motion predictor 116 supplies the RD cost (P) which is the smallest RD
25 cost among the RD costs of the combinations of the inter prediction mode and the angular index and the corresponding motion vector and inter prediction mode information to the motion compensator 113. [0129]
30 In this way, the motion compensator 113 performs a compensation process on the reference image supplied from
38

SP307551WO00
the frame memory 112 based on the motion vector and the inter prediction mode information supplied from the motion predictor 116 and generates a prediction image. Moreover, the motion compensator 113 supplies the RD cost 5 (P) supplied from the motion predictor 116 and the generated prediction image to the selector 117. [0130]
Fig. 10 is a flowchart for describing the details of the process of step S36 of Fig. 9. 10 [0131]
In step S51 of Fig. 10, the orthogonal transformer 104 determines whether the block size of the inter
prediction is 4x4 pixels. [0132]
15 When it is determined in step S51 that the block size of the inter prediction is 4x4 pixels, in step S52, the orthogonal transformer 104 performs a ROTxDCT process according to an angular index. Specifically, the 4x4 DCTxROT 501 (Fig. 5) of the orthogonal transformer 104
2 0 performs a combined transform of DCT and ROT on the
residual information supplied from the computing device 103 according to the angular index set in step S35 of Fig. 9. The 4x4 DCTxROT 501 supplies the coefficients obtained as a result of the transform to the 4x4 Quant
25 419 of the quantizer 105.
[0133] -■ . . .
When it is determined in step S51 that the block size of the inter prediction is not 4x4 pixels, in step S53, the orthogonal transformer 104 performs a DCT
30 process which is a process of performing DCT on the
residual information supplied from the computing device
39

SP307551WO00
103. Specifically, the 8x8 DCT 412, the 16x16 DCT 413, the 32x32 DCT 414, the 64x64 DCT 415, or the 128x128 DCT 416 corresponding to the block size of the inter prediction of the orthogonal transformer 104 performs DCT 5 on the residual information. The 8x8 pixel low frequency component among the coefficients obtained as a result of the DCT is supplied to the 8x8 ROT 418, and the remaining high frequency component is supplied to the 16x16 Quant 421, the 32x32 Quant 422, the 64x64 Quant 423, or the 10 128x128 Quant 424 corresponding to the block size of the inter prediction. [0134]
In step S54, the 8x8 ROT 418 of the orthogonal transformer 104 performs a ROT process according to the 15 angular index set in step S35 of Fig. 9 with respect to the 8x8 pixel (8x8 size) coefficients of the low frequency component. The 8x8 ROT 418 supplies the 8x8 pixel coefficients obtained as a result of the ROT process to the 8x8 Quant 420, the 16x16 Quant 421, the 20 32x32 Quant 422, the 64x64 Quant 423, or the 128x128
Quant 424 corresponding to the block size of the intra prediction. [0135]
Fig. 11 is a flowchart for describing the process 25 of step S40 of Fig. 9 in detail. [0136]
In step S71 of Fig. 11, the inverse orthogonal transformer 109 determines whether the block size of the inter prediction is 4x4 pixels. 30 [0137]
When it is determined in step S71 that the block
40

SP307551WO00
size of the inter prediction is 4x4 pixels, in step S72, the inverse orthogonal transformer 109 performs an inverse ROTxDCT process according to the angular index. Specifically, the 4x4 Inv ROTxInv DCT 502 (Fig. 5) of the 5 inverse orthogonal transformer 109 performs a combined transform of inverse ROT and inverse DCT on the coefficients supplied from the 4x4 Inv Quant 451 of the dequantizer 108 according to the angular index set in step S35 of Fig. 9. The 4x4 Inv ROTxInv DCT 502 supplies
10 the residual information obtained as a result of the transform to the adder 110. [0138]
When it is determined in step S71 that the block size of the inter prediction is not 4x4 pixels, the flow
15 proceeds to step S73. In step S73, the 8x8 Inv ROT 458 (Fig. 7) of the inverse orthogonal transformer 109 performs an inverse ROT process which is a process of performing inverse ROT according to the angular index set in step S35 of Fig. 9 with respect to the 8x8 pixel (8x8
20 size) coefficients of the low frequency component among the coefficients of a size of 8x8 pixels or larger supplied from the dequantizer 108. The 8x8 Inv ROT 458 supplies the coefficients obtained as a result of the inverse ROT process to the 8x8 Inv DCT 460, the 16x16 Inv
25 DCT 461, the 32x32 Inv DCT 462, the 64x64 Inv DCT 463, or the 128x128 Inv DCT 464 corresponding to the block size of the inter prediction. [0139]
In step S74, the 8x8 Inv DCT 460, the 16x16 Inv DCT
30 461, the 32x32 Inv DCT 462, the 64x64 Inv DCT 463, or the 128x128 Inv DCT 464 of the inverse orthogonal transformer
41

SP307551WO00
109 performs an inverse DCT process which is a process of performing inverse DCT on the coefficients supplied from the 8x8 Inv ROT 458 and the coefficients supplied from the dequantizer 108. The residual information obtained 5 as a result of the inverse DCT process is supplied to the adder 110. [0140]
Fig. 12 is a flowchart for describing a process of calculating the RD cost (I) when the intra prediction of
10 step S12 of Fig. 8 is used in detail. [0141]
In step SlOl of Fig. 12, the intra predictor 114 sets the block size of the intra prediction to one which has not been set among the 4x4, 8x8, 16x16, 32x32, 64x64,
15 and 128x128 pixels. [0142]
In step S102, the intra predictor 114 sets an intra prediction mode (Intra direction mode) to one which has not been set among the intra direction modes of which the
20 intra direction mode number is 0, 1, 2, 3, 4, 5, 6, 7, or 8. The intra direction mode number is a number unique to the intra prediction mode, and in the present embodiment, eight intra prediction modes of the numbers 0 to 8 are prepared.
25 [0143]
In step S103, the intra predictor 114 performs motion prediction with the block size and the intra prediction mode set in step SlOl. Specifically, the intra predictor 114 performs an intra prediction process
30 in the set intra prediction mode in respective blocks of the block size set in step SlOl using the image supplied
42

SP307551WO00
from the frame rearrangement buffer 102 and the reference image supplied from the adder 110 and generates the prediction image. The intra predictor 114 supplies the generated prediction image to the computing device 103 5 via the selector 117. [0144]
In step S104, the computing device.103 computes a difference between the image corresponding to the input signal and the intra prediction image (the prediction
10 image generated by the intra prediction process). The
computing device 103 supplies the difference obtained as a result of the computation to the orthogonal transformer 104 as residual information. [0145]
15 The processes of steps S105 to SllO are the same as the processes of steps S35 to S40 of Fig. 9, and description thereof will not be provided. [0146]
After the process of step SllO is performed, the
20 flow returns to step S105, and the processes of steps S105 to SllO are repeatedly performed until all of the angular indices of the index numbers 0 to 3 are set as the angular index. Moreover, when all of the angular indices of the index numbers 0 to 3 are set as the
25 angular index, the flow returns to step S102. Moreover, the processes of steps S102 to SllO are repeatedly performed until all intra prediction modes of the intra direction mode niombers 0 to 8 are set as the intra prediction mode.
30 [0147]
Moreover, when all of the intra direction mode
43

SP307551WO00
numbers 0 to 8 are set as the intra prediction mode, the flow returns to step SlOl. Moreover, the processes of steps SlOl to SllO are repeatedly performed until all sizes of the 4x4, 8x8, 16x16, 32x32, 64x64, and 128x128 5 pixels are set as the block size of the intra prediction. [0148]
Moreover, when all sizes of the 4x4, 8x8, 16x16, 32x32, 64x64, and 128x128 pixels are set as the block size of the intra prediction, all of the angular indices
10 of the index niimbers 0 to 3 are set as the angular index with respect to the block of each block size, and when all of the intra prediction modes of the intra prediction modes 0 to 8 are set as the intra prediction mode, the flow proceeds to step Sill.
15 [0149]
In step Sill, the intra predictor 114 computes a RD cost from the quantized code information and the decoded image with respect to each combination of the intra prediction block size, the intra prediction mode, and the
20 angular index. Specifically, the intra predictor 114 generates a prediction image using the reference image supplied from the frame memory 112 with respect to each combination of the intra prediction block size, the intra prediction mode, and the angular index. Moreover, the
25 intra predictor 114 computes a difference between the prediction image and the image supplied from the frame rearrangement buffer 102. Moreover, the motion predictor 116 coiaputes Expression (1) described above and calculates the RD cost using the difference, the
30 occurrence coding amount of the compressed image obtained by the process of step 3108, and the like.
44

SP307551WO00
[0150]
Moreover, the intra predictor 114 uses the smallest RD cost among the RD costs of the respective combinations of the intra prediction block size, the intra prediction 5 mode, and the angular index as the RD cost (I). That is, the intra predictor 114 supplies the RD cost (I) which is the smallest RD cost among the RD costs of the respective combinations of the intra prediction block size, the intra prediction mode, and the angular index and the
10 corresponding prediction image to the selector 117. [0151]
Figs. 13, 14, and 15 are flowcharts of the processes of the decoder of Fig. 3. [0152]
15 Fig. 13 is a flowchart for describing a macroblock (MB) decoding process. [0153]
In step S121 of Fig. 13, the lossless decoder 217 reads and acquires the image compression information of
20 the current macroblock from the storage buffer 216 and losslessly decodes the image compression information according to a scheme corresponding to the lossless encoding scheme of the lossless encoder 106 of Fig. 2. By this lossless decoding, the intra prediction mode
25 information or the inter prediction mode information is extracted as the information that indicates the optimal prediction mode of the current macroblock. [0154]
In step S122, the lossless decoder 217 determines
30 whether the information that indicates the optimal prediction mode extracted in step S121 is the intra
45

SP307551WO00
prediction mode information. When it is determined in step S122 that the information is the intra prediction mode information, in step S123, the decoder decodes the current macroblock (the MB) according to intra prediction. 5 The details of the process of step S123 will be described with reference to Fig. 15 described later. [0155]
On the other than, when it is determined in step S122 that the information is not the intra prediction
10 mode information, that is, the information that indicates the optimal prediction mode extracted in step S121 is inter prediction mode information, the flow proceeds to step S124. [0156]
15 In step S124, the decoder decodes the current
macroblock according to inter prediction. The details of the process of step S124 will be described with reference to Fig. 14 described later. [0157]
20 Fig. 14 is a flowchart for describing the details of the process of step S124 of Fig. 13. [0158]
In step S141 of Fig. 14, the lossless decoder 217 extracts the quantized coefficients corresponding to the
25 inter prediction block size, the motion vector (MV), the angular index information, and the residual information ■ (difference information) from the image compression information (stream information) acquired from the storage buffer 216. Specifically, the lossless decoder
30 2l7 losslessly decodes the image compression information to obtain the inter prediction mode information, the
46

SP307551WO00
motion vector, the angular index information, and the quantized coefficients. Moreover, the lossless decoder
217 recognizes the block size of the inter prediction
corresponding to the inter prediction mode information.
5 The lossless decoder 217 supplies the quantized
coefficients to the dequantizer 218 in respective blocks of the block size corresponding to the inter prediction mode information. Moreover, the lossless decoder 217 supplies the inter prediction mode information and the
10 motion vector to the motion compensator 224 and supplies the angular index to the inverse orthogonal transformer 219. [0159]
In step S142, the motion compensator 224 performs a
15 motion compensation process (MC process) on the reference image supplied from the frame memory 223 according to the inter prediction mode information and the motion vector supplied from the lossless decoder 217. Moreover, the motion compensator 224 supplies the prediction image
20 obtained as a result of the motion compensation process to the adder 220 via the switch 227. [0160]
In step S143, the dequantizer 218 performs a dequantization process on the quantized coefficients
25 supplied from the lossless decoder 217. Specifically, the 4x4 Inv Quant 601, the 8x8 Inv Quant-602, the 16x16 Inv Quant 603, the 32x32 Inv Quant 604, the 64x64 Inv Quant 605, or the 128x128 Inv Quant 606 corresponding to the inter prediction block size of the dequantizer 218
30 dequantizes the quantized coefficients. The dequantizer
218 supplies the coefficients obtained as a result of the
47

SP307551WO00
dequantization process to the inverse orthogonal
transformer 219.
[0161]
In step S144, the inverse orthogonal transformer 5 219 performs an inverse ROT process or the like according to the angular index supplied from the lossless decoder 217 with respect to the coefficients corresponding to the difference information (the residual information) supplied from the dequantizer 218. Since the details of
10 the process of step S144 are the same as those described in Fig. 11, the description thereof will not be provided. [0162]
In step S145, the adder 22 0 adds the residual information (inverse ROT information) obtained as a
15 result of the process of step S144 to the prediction image (prediction signal) supplied from the motion compensator 224 via the switch 227 to obtain a decoded image. The decoded image is supplied to the intra predictor 225, is supplied to the frame memory 223 via
20 the deblocking filter 226, or is supplied to the outside via the deblocking filter 22 6, the frame rearrangement buffer 221, and the D/A converter 222. [0163]
Fig. 15 is a flowchart for describing the details
25 of the process of step S123 of Fig. 13. [0164]
In step S161 of Fig. 15, the lossless decoder 217 extracts quantized coefficients corresponding to the intra prediction block size, the intra prediction mode,
30 the angular index information, and the residual
information (difference information) from the image
48

SP307551WO00
compression information (stream information) acquired from the storage buffer 216. Specifically, the lossless decoder 217 losslessly decodes the image compression information to obtain the intra prediction mode 5 information, the angular index information, and the
quantized coefficients. Moreover, the lossless decoder 217 recognizes the intra prediction mode and the intra prediction block size from the intra prediction mode information. The lossless decoder 217 supplies the
10 quantized coefficients to the dequantizer 218 in
respective blocks of the intra prediction block size. Moreover, the lossless decoder 217 supplies the intra prediction mode information to the intra predictor 225 and supplies the angular index to the inverse orthogonal
15 transformer 219. [0165]
In step SI62, the intra predictor 225 performs an intra prediction process on the reference image supplied from the adder 220 according to the intra prediction mode
20 information supplied from the lossless decoder 217.
Moreover, the intra predictor 225 supplies the prediction image obtained as a result of the intra prediction process to the adder 220 via the switch 227. [0166]
25 In step S163, the dequantizer 218 performs a
dequantization process on the quantized coefficients supplied from the lossless decoder 217 in a manner similarly to the process of step S143 of Fig. 14. The dequantizer 218 supplies the coefficients obtained as a
30 result of the dequantization process to the inverse orthogonal transformer-219.
49

SP307551WO00
[0167]
In step SI64, the inverse orthogonal transformer
219 performs an inverse ROT process or the like on the
coefficients corresponding to the difference information 5 supplied from the dequantizer 218 in a manner similarly
to the process of step S144 according to the angular
index supplied from the lossless decoder 217.
[0168]
In step S165, the adder 220 adds the residual 10 information (inverse ROT information) obtained as a
result of the process of step S164 to the prediction
image (prediction signal) supplied from the intra
predictor 225 via the switch 227 to obtain the decoded
image. The decoded image is supplied to the intra 15 predictor 225, is supplied to the frame memory 223 via
the deblocking filter 22 6, or is output to the outside
via the deblocking filter 22 6, the frame rearrangement
buffer 221, and the D/A converter 222.
[0169] 20 [Description of Computer to which Present technology is
applied]
[0170]
Next, the above-described series of processing can
be executed not only by hardware but also by software. 25 When the series of processing is executed by software, a
program included in the.software is installed -in-a .
general-purpose computer or the like.
[0171]
With reference now to Fig. 16, an exemplary 30 configuration of a computer according to an embodiment of
the present technology, in which a program for executing
50

SP307551WO00
the above-described series of processing is installed,
will be described.
[0172]
The program may be preliminarily recorded in a hard 5 disk 705 or a ROM 703 as a recording medium equipped in
the computer.
[0173]
Alternatively, the program may be stored (recorded)
in a removable recording medium 711. The removable 10 recording medium 711 may be provided as so-called package
software. Here, the removable recording medium 711 may
be, for example, a flexible disk, a CD-ROM (compact disc
read only memory), a MO (magneto optical) disc, a DVD
(digital versatile disc), a magnetic disc, a 15 semiconductor memory or the like.
[0174]
The program may be installed in the internal hard
disk 7 05 by down loading the program to a computer via a
communication network or a broadcasting network, in 20 addition to installing the program in the computer from
the removable recording medium 711 as described above.
That is to say, the program may be transferred in a
wireless manner from a download site to the computer via
a digital broadcasting satellite or may be transferred in 25 a wired manner to the computer via a network such as a
LAN (local area network),.or the Internet.
[0175}
The computer has incorporated therein a CPU
(central processing unit) 7 02, and an input/output 30 interface 710 is connected to the CPU 702 via a bus 701.
[0176]
51

SP307551WO00
The CPU 702 executes the program stored in the ROM (read only memory) 703 in response to commands which are input via the input/output interface 710 by a user operating an input unit 707 or the like. Alternatively, 5 the CPU 702 executes the program stored in the hard disk 705 by loading the program in a RAM (random access memory) 704. [0177]
In this way, the CPU 7 02 executes the processing
10 corresponding to the above-described flowcharts or the
processing performed by the configuration illustrated in the block diagrams. Then, the CPU 702 outputs, transmits, or records the processing results through an output unit 705, through a communication unit 708, or in the hard
15 disk 705, for example, via the input/output interface 710 as required. [0178]
The input unit 7 07 includes a keyboard, a mouse, a microphone, and the like. The output unit 706 includes
20 an LCD (liquid crystal display), a speaker, and the like. [0179]
Here, in this specification, the processing that the computer executes in accordance with the program may not be executed in a time-sequential manner in the order
25 described in the flowcharts. That is to say, the
processing that the computer executes-in accordance with the program includes processing that is executed in parallel or separately (for example, parallel processing or object-based processing).
30 [0180]
Moreover, the program may be executed by a single
52

•

SP307551WO00

computer (processor) and may be executed by a plurality of computers in a distributed manner. Furthermore, the program may be executed by being transferred to a computer at a remote location. 5 [0181]
The embodiments of the present technology are not limited to the above-described embodiments, and various changes can be made without departing from the spirit of the present technology.
10 [0182]
Moreover, the present technology can take the following configurations. [0183]
(1) An image processing device including:
15 'a dequantization unit that dequantizes a quantized image to obtain a low frequency component having a predetermined size of the image, which is obtained by performing a second orthogonal transform after a first orthogonal transform, and to obtain a high frequency
20 component, which is a component other than the low
frequency component of the image and is obtained by the first orthogonal transform; and
an inverse orthogonal transform unit that, when a size of the image is the predetermined size, performs a
25 third inverse orthogonal transform, which is a combined transform of a first inverse orthogonal transform corresponding to the first orthogonal transform and a second inverse orthogonal transform corresponding to the second orthogonal transform, on the image which is the
30 low frequency component, and that, when the size of the
image is larger than the predetermined size, performs the
53

SP307551WO00
second inverse orthogonal transform on the low frequency component and performs the first inverse orthogonal transform on the low frequency component having been subjected to the second inverse orthogonal transform and 5 the high frequency component obtained by the dequantization unit.
(2) The image processing device according to (1), wherein
the predetermined size is 4x4 pixels.
10 (3) The image processing device according to (1), wherein
the predetermined size is 4x4 pixels when the size of the image is 4x4 pixels and is 8x8 pixels when the size of the image is 8x8 pixels or lager,
15 when the size of the image is 4x4 pixels, the inverse orthogonal transform unit performs the third inverse orthogonal transform on the image which is the low frequency component, when the size of the image is 8x8 pixels or larger, the inverse orthogonal transform
20 unit performs the second inverse orthogonal transform on the low frequency component and performs the first inverse orthogonal transform on the low frequency component having been subjected to the second inverse orthogonal transform and the high frequency component
25 obtained by the dequantization unit.
(4) The image processing device according to any one of (1) to (3), wherein
the first orthogonal transform is a discrete cosine transform (DCT), and
30 the second orthogonal transform is a rotation transform (ROT).
54

SP307551WO00
(5) The image processing device according to any
one of (1) to (4), further including:
an orthogonal transform unit that, when the size of the image is the predetermined size, performs a third 5 orthogonal transform, which is a combined transform of the first orthogonal transform and the second orthogonal transform, on the image, and that, when the size of the image is larger than the predetermined size, performs the first orthogonal transform on the image and performs the
10 second orthogonal transform on the low frequency
component having the predetermined size of the image having been subjected to the first orthogonal transform; and
a quantization unit that quantizes the image having
15 the predetermined size having been subjected to the third orthogonal transform or quantizes the high frequency component, which is the component other than the low frequency component and is obtained by the first orthogonal transform, and the low frequency component
20 obtained by the second orthogonal transform.
(6) An image processing method of an image
processing device including:
a dequantization unit that dequantizes a quantized image to obtain a low frequency component having a 25 predetermined size of the image, which is obtained by performing a second orthogonal transform after a first orthogonal transform, and to obtain a high frequency component, which is a component other than the low frequency component of the image and is obtained by the 30 first orthogonal transform; and
an inverse orthogonal transform unit that, when a
55

SP307551WO00
size of the image is the predetermined size, performs a third inverse orthogonal transform, which is a combined transform of a first inverse orthogonal transform corresponding to the first orthogonal transform and a 5 second inverse orthogonal transform corresponding to the second orthogonal transform, on the image which is the low frequency component, and that, when the size of the image is larger than the predetermined size, performs the second inverse orthogonal transform on the low frequency
10 component and performs the first inverse orthogonal transform on the low frequency component having been subjected to the second inverse orthogonal transform and the high frequency component obtained by the dequantization unit,
15 the method including the steps of:
allowing the dequantization unit to obtain the low frequency component and the high frequency component; and
allowing the inverse orthogonal transform unit to perform the third inverse orthogonal transform on the
20 image which is the low frequency component when the size of the image is the predetermined size, to perform the second inverse orthogonal transform on the low frequency component when the size of the image is larger than the predetermined size, and to perform the first inverse
25 orthogonal transform on the low frequency component
having been subjected to the second inverse orthogonal transform and the high frequency component obtained by the dequantization unit.
(7) A program for causing a computer to function
30 as:
a dequantization unit that dequantizes a quantized
56

SP307551WO00
image to obtain a low frequency component having a predetermined size of the image, which is obtained by performing a second orthogonal transform after a first orthogonal transform, and to obtain a high frequency 5 component, which is a component other than the low
frequency component of the image and is obtained by the first orthogonal transform; and
an inverse orthogonal transform unit that, when a size of the image is the predetermined size, performs a
10 third inverse orthogonal transform, which is a combined transform of a first inverse orthogonal transform corresponding to the first orthogonal transform and a second inverse orthogonal transform corresponding to the second orthogonal transform, on the image which is the
15 low frequency component, and that, when the size of the
image is larger than the predetermined size, performs the second inverse orthogonal transform on the low frequency component and performs the first inverse orthogonal transform on the low frequency component having been
20 subjected to the second inverse orthogonal transform and the high frequency component obtained by the dequantization unit.
REFERENCE SIGNS LIST 25 [0184]
104 Orthogonal transformer
105 Quantizer

108 Dequantizer
109 Inverse orthogonal transformer 30 218 Dequantizer
219 Inverse orthogonal transformer
57

701 ■ Bus
7 02 CPU
703 ROM
704 RAM
5 705 Hard disk
706 Output unit
7 07 Input unit
708 Communication unit
709 Drive
10 710 Input/output interface
711 Removable recording medium
58

SP307551WO00

CLAIMS
1. An image processing device comprising:
a dequantization unit that dequantizes a quantized
5 image to obtain a low frequency component having a
predetermined size of the image, which is obtained by
performing a second orthogonal transform after a first
orthogonal transform, and to obtain a high frequency
component, which is a component other than the low
10 frequency component of the image and is obtained by the
first orthogonal transform; and
an inverse orthogonal transform unit that, when a
size of the image is the predetermined size, performs a
third inverse orthogonal transform, which is a combined
15 transform of a first inverse orthogonal transform
corresponding to the first orthogonal transform and a
second inverse orthogonal transform corresponding to the
second orthogonal transform, on the image which is the
low frequency component, and that, when the size of the
20 image is larger than the predetermined size, performs the
second inverse orthogonal transform on the low frequency
component and performs the first inverse orthogonal
transform on the low frequency component having been
subjected to the second inverse orthogonal transform and
25 the high frequency component obtained by the
dequantization unit.
2. The image processing device according to claim 1,
wherein
30 the predetermined size is 4x4 pixels.
59
SP307551WO00
t
3. The image processing device according to claim 1,
wherein
the predetermined size is 4x4 pixels when the size
of the image is 4x4 pixels and is 8x8 pixels when the
5 size of the image is 8x8 pixels or larger,
when the size of the image is 4x4 pixels, the
inverse orthogonal transform unit performs the third
inverse orthogonal transform on the image which is the
low frequency component, and
10 when the size of the image is 8x8 pixels or larger,
the inverse orthogonal transform unit performs the second
inverse orthogonal transform on the low frequency
component and performs the first inverse orthogonal
transform on the low frequency component having been
15 subjected to the second inverse orthogonal transform and
the high frequency component obtained by the
dequantization unit.
4. The image processing device according to claim 1,
20 wherein
the first orthogonal transform is a discrete cosine
transform (DCT), and
the second orthogonal transform is a rotation
transform (ROT).
25
5. The image processing device according to claim 1,
further comprising:
an orthogonal transform unit that, when the size of
the image is the predetermined size, performs a third
30 orthogonal transform, which is a combined transform of
the first orthogonal transform and the second orthogonal
60
SP307551WO00
transform, on the image, and that, when the size of the
image is larger than the predetermined size, performs the
first orthogonal transform on the image and performs the
second orthogonal transform on the low frequency
5 component having the predetermined size of the image
having been subjected to the first orthogonal transform;
and
a quantization unit that quantizes the image having
the predetermined size having been subjected to the third
10 orthogonal transform or quantizes the high frequency
component, which is the component other than the low
frequency component and is obtained by the first
orthogonal transform, and the low frequency component
obtained by the second orthogonal transform.
15
6. An image processing method of an image processing
device comprising:
a dequantization unit that dequantizes a quantized
image to obtain a low frequency component having a
20 predetermined size of the image which is obtained by
performing a second orthogonal transform after a first
orthogonal transform and to obtain a high frequency
component which is a component other than the low
frequency component of the image and is obtained by the
25 first orthogonal transform; and
an inverse orthogonal transform unit that, when a
size of the image is the predetermined size, performs a
third inverse orthogonal transform, which is a combined
transform of a first inverse orthogonal transform
30 corresponding to the first orthogonal transform and a
second inverse orthogonal transform corresponding to the
61
SP307551WO00
•
second orthogonal transform, on the image which is the
low frequency component, and that, when the size of the
image is larger than the predetermined size, performs the
second inverse orthogonal transform on the low frequency
5 component and performs the first inverse orthogonal
transform on the low frequency component having been
subjected to the second inverse orthogonal transform and
the high frequency component obtained by the
dequantization unit,
10 the method comprising the steps of:
allowing the dequantization unit to obtain the low
frequency component and the high frequency component; and
allowing the inverse orthogonal transform unit to
perform the third inverse orthogonal transform on the
15 image which is the low frequency component when the size
of the image is the predetermined size, to perform the
second inverse orthogonal transform on the low frequency
component when the size of the image is larger than the
predetermined size, and to perform the first inverse
20 orthogonal transform on the low frequency component
having been subjected to the second inverse orthogonal
transform and the high frequency component obtained by
the dequantization unit.
25 7. A program for causing a computer to function as:
a dequantization unit that dequantizes a quantized •
image to obtain a low frequency component having a
predetermined size of the image, which is obtained by
performing a second orthogonal transform after a first
30 orthogonal transform, and to obtain a high frequency
component, which is a component other than the low
62
SP307551WO00
frequency component of the image and is obtained" by the
first orthogonal transform; and
an inverse orthogonal transform unit that, when a
size of the image is the predetermined size, performs a
5 third inverse orthogonal transform, which is a combined
transform of a first inverse orthogonal transform
corresponding to the first orthogonal transform and a .
second inverse orthogonal transfotm corresponding to the
second orthogonal transform, on the image which is the
10 low frequency component, and that, when tke size of the
image is larger than the predetermined size, performs the
second inverse orthogonal transform on the Ipw frequency
component and performs the first inverse orthogonal
I
transform on the low frequency component having been
15 subjected to the second inverse orthogonal transform and
the high frequency component obtained by the -^
dequantization unit.