JP2888292B2 - Two-dimensional inverse discrete cosine transform (IDCT) device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a two-dimensional inverse discrete cosine transform (IDCT) device used in video decoding.

[0002]

2. Description of the Related Art In recent years, microprocessors have adopted an operation instruction using a divided ALU system for improving the operation capability by processing data in parallel. The split ALU instruction is, for example, an instruction that utilizes a 64-bit ALU as four 16-bit ALUs and processes four pieces of data in parallel. In the case of implementing an 8-point × 8-point two-dimensional IDCT, as shown in FIG. 14, first, an 8-point one-dimensional IDCT in the row direction represented by row0 to row7 is calculated by using a DCT coefficient as an input, and the result is used as an input.
8-point one-dimensional IDC in the column direction represented by 0 to column7
A method called Row-Column method for calculating T is generally used. Row-Co using split ALU instruction
FIG. 15 shows a method of calculating the lum method. To read four consecutive data and use the divided ALU instruction, the DCT coefficients are transposed in advance. As the operation in the row direction, R0 is an 8-point one-dimensional IDC from row0 to row3.
Four T's are calculated at the same time, and four one-dimensional IDCTs from row 4 to row 7 are simultaneously calculated at R1. In the column direction operation, a value obtained by transposing the operation results of R0 and R1 is input and C
0, 8 points 1-dimensional I from column0 to column3
Four DCTs are simultaneously calculated, and four eight-point one-dimensional IDCTs from column4 to column7 are simultaneously calculated in C1. As described above, when the operation is performed in four parallels using the divided ALU instruction, an eight-point × eight-point two-dimensional IDCT can be realized by calculating the eight-point one-dimensional IDCT four times.

In order to further speed up the 8-point × 8-point two-dimensional IDCT operation, a method has been proposed in which the operation with zero input is omitted. An example is the 1996 ICI
Vol. 2, pp. 153 to 156 (ICIP '96, V
ol. 2, pp. 153-156).
FIG. 21 shows a configuration of a conventional example. Effective coefficient detector 2100
Are four DCTs that take DCT coefficients as input and process simultaneously
Checks if coefficients contain non-zero data. I
DCT scheme selection section 2101 includes effective coefficient detection section 2100
Complete IDC that performs the operation without omitting the result of
One of the T operation method and the partial IDCT operation method in which the operation in which the DCT coefficient is zero is omitted is selected. Switch 21
02 and the selector 2104 are provided by an IDCT method selection unit 2101.
Switch according to the selection result of. IDCT operation unit 210
3 is the IDCT selected by the IDCT scheme selection unit 2101
The IDCT operation is performed using the operation method. The IDCT operation result is output through the selector 2104.

[0004] Complete ID used in IDCT calculation section 2103
The IDCT operation method shown in FIG. 15 is used as the CT operation method. Also, the part I used in the IDCT operation unit 2103
The partial IDCT operation method 2 shown in FIG. 17 is used as the DCT operation method. When the input DCT coefficients are collectively represented by X0 to X15 for each of the four DCT coefficients processed in parallel as shown in FIG. 5, the partial IDCT operation method 2 is 1700
This is an IDCT calculation method used when a non-zero coefficient exists in X0 to X3 and all of X4 to X15 are zero. Since the input data in the row direction exists only in X0 to X3,
In R0, an IDCT operation in which the operations from X4 to X7 are omitted is performed. R1 does not perform any operation because the input data is all zero. The operation result in the row direction is X0 'to X7' of 1701.
Only non-zero and all others are zero. When the transposition in the xy direction is performed, the input data of the IDCT in the column direction is non-zero in only Y0 to Y3 and Y8 to Y11 in 1702, and all others are zero. In C0, an IDCT operation in which operations other than Y0 to Y3 are omitted is performed. In C1, an IDCT operation in which operations other than Y8 to Y11 are omitted is performed, and 8 points × 8 points 2
A dimension IDCT operation result 1703 is obtained.

[0005] Partial ID used in IDCT calculation section 2103
As the CT operation method, the partial IDCT operation method 3 shown in FIG. 18 may be used. The partial IDCT operation method 3 is 180
There are non-zero coefficients at X0 to X7 of 0, and at X8 to X15
Is an IDCT operation method used when all are zero.
Since input data in the row direction exists only in X0 to X7, an 8-point one-dimensional IDCT operation without omission is performed in R0. R1 does not perform any operation because the input data is all zero. As for the operation result in the row direction, only X0 'to X7' of 1801 are non-zero, and all others are zero. When the transposition in the xy direction is performed, the input data of the IDCT in the column direction becomes Y0 of 1802.
To Y3 and Y8 to Y11 are non-zero, and all others are zero. In C0, I in which operations other than Y0 to Y3 are omitted
A DCT operation is performed, and an IDCT operation in which the operations other than Y8 to Y11 are omitted is performed in C1 to obtain an 8-point × 8-point two-dimensional IDCT operation result 1803.

The selection method performed by the IDCT method selection unit 2101 is based on the partial IDC prepared in the IDCT operation unit 2103.
Determined according to the T operation method. For example, when the above partial IDCT operation 2 is used as the partial IDCT operation method, X4
If X15 to X15 are all zero, the partial IDCT operation method is selected, and otherwise the complete IDCT operation method is selected. When the partial IDCT operation method 3 is used as the partial IDCT operation method, if X8 to X15 are all zero, the partial IDCT operation method is selected.
Select the DCT operation method.

[0007]

[Problems to be Solved by the Invention] Partial IDCT to be prepared
If there is one calculation method, the effect of reducing the calculation amount differs depending on what partial IDCT calculation method is prepared. For example, when the partial IDCT operation method 2 is used, the partial IDCT operation method can be selected in 30% of the encoded blocks, and when the partial IDCT operation method 3 is used, the partial IDCT operation can be selected in 60% of the encoded blocks. For an image in which a method can be selected, preparing the partial IDCT calculation method 3 is more effective in reducing the amount of calculation.
For an image in which the selectivity of the T operation is 30%, it is better to prepare the partial IDCT operation method 2 in which the amount of reduction in the operation is large. As described above, the effect of reducing the amount of operation of the partial IDCT operation and the complete IDCT operation is a trade-off between the selectivity of the partial IDCT operation and the operation amount that can be reduced by using the partial IDCT operation.
The selectivity of the partial IDCT operation depends on the distribution of non-zero coefficients in DCT coefficients. The distribution of non-zero coefficients depends on the image, and the I-picture type for performing intra-frame encoding without using motion compensation prediction, and the P-type for performing forward prediction encoding that refers only to an image signal input earlier in time. According to each picture type, a B picture type for performing bidirectional predictive encoding with reference to a picture type and a forward and backward image signal,
In addition, since it changes according to each macro block type of the INTRA block type for performing intra-frame coding and the INTER block type for performing inter-frame coding, only one partial IDCT calculation method is prepared, thereby reducing the calculation amount for all frames. It is difficult to get an effect. In consideration of the above problem, if a plurality of partial IDCT operation methods are prepared so that the effect of reducing the operation amount can be obtained regardless of the distribution of DCT coefficients, the operation amount due to the selection of the IDCT operation method increases. Occurs.

There is also a problem that the amount of calculation required to detect the effective coefficient in the effective coefficient detection unit 2100 is large. The reason is that the effective coefficient detection unit examines all DCT coefficients and determines whether or not a non-zero coefficient exists.

A first object of the present invention is to provide an effective coefficient detection result when a plurality of partial IDCT operation methods are prepared so that the effect of reducing the operation amount can be obtained regardless of the distribution of non-zero coefficients in DCT coefficients. In order to reduce the amount of calculation, the IDCT method selection unit that selects the IDCT calculation method that can reduce the amount of calculation the most.

A second object of the present invention is to reduce the amount of calculation required for detecting an effective coefficient.

[0011]

An eight-point by eight-point two-dimensional IDCT apparatus according to the present invention has a plurality of IDCT scheme selection units.
It has a switch and a selector for switching it according to the picture type or macroblock type. IDC
The T-method selector selects a selection candidate of several IDCT calculation methods in advance according to the characteristic of the DCT coefficient from a plurality of IDCT calculation methods prepared by the IDCT calculation unit for each picture type or macroblock type. Based on the effective coefficient detection result, an IDCT operation method that requires the least amount of operation is selected from the selected IDCT operation methods.

Another feature is that effective coefficient detection is performed on the assumption that data up to the EOB code is non-zero data.

According to the present invention, a processor that executes the same processing on a plurality of data with one instruction is used.
An 8 × 8-point two-dimensional IDCT apparatus that performs point-one-dimensional inverse discrete cosine transform (IDCT) in four parallels; an effective coefficient detection unit that determines whether four DCT coefficients to be processed simultaneously are non-zero; Based on the effective coefficient detection result, the first and the first to select the IDCT operation method with the least operation amount
A second and a third IDCT scheme selector, a first switch and a first selector for switching the IDCT scheme selector according to a picture type, an IDCT calculator having a plurality of IDCT schemes, and the IDCT scheme selector Has a second switch and a second selector for switching an IDCT operation method used in the IDCT operation unit according to the IDCT operation method selected in the above. The IDCT operation unit includes an I type for performing intra-frame encoding, Several IDCT operation methods are prepared in advance according to the characteristics of the distribution of the DCT coefficients in each of the P type for performing directional prediction encoding and the B type for performing bidirectional prediction encoding. Switching the first switch and the first selector, and if the picture type is an I picture, the first ID The T method selector, the second IDCT method selector for a P picture, and the third IDCT method selector for a B picture use the IDCT operation method that can reduce the amount of operation most in accordance with the effective coefficient detection result. And switching the second switch and the second selector in accordance with the IDCT operation method to select the IDC
IDC by the IDCT operation method selected by the T method selection unit
8 points × 8 points 2D IDC characterized by performing T operation
A T device is obtained.

According to the present invention, an eight-point one-dimensional inverse discrete cosine transform (IDCT) is executed in four parallel by using a processor that executes the same processing on a plurality of data with one instruction. In an eight-point two-dimensional IDCT apparatus, an effective coefficient detection unit that determines whether or not four DCT coefficients to be simultaneously processed are non-zero, and an IDCT operation method that requires the least amount of operation is selected based on the effective coefficient detection result. A first and a second IDCT scheme selection section, a first switch and a first selector for switching the IDCT scheme selection section according to macroblock type, an IDCT computation section having a plurality of IDCT computation schemes, and an IDCT scheme selection section A second switch and a second selector for switching an IDCT operation method used in the IDCT operation unit according to the IDCT operation method selected by the unit. The IDCT operation unit includes several IDCT operation methods in advance according to the characteristics of the distribution of DCT coefficients in each of the macro block types of the INTRA type for performing intra-frame coding and the INTER type for performing inter-frame coding. Are prepared, and the first switch and the first switch are set according to a macroblock type.
Switch the macro block type to INT
If RA, the first IDCT scheme selector, INTER
Then, in the second IDCT scheme selection unit, the ID that can reduce the amount of calculation most according to the effective coefficient detection result
A CT operation method is selected, the second switch and the second selector are switched according to the IDCT operation method, and the IDCT operation is performed by the IDCT operation method selected by the IDCT method selection unit. A point × 8 point two-dimensional IDCT device is obtained.

[0015] The IDCT scheme selection unit preliminarily limits the candidates of the IDCT scheme according to each picture type or macroblock type from among a plurality of IDCT schemes prepared by the IDCT computation unit. For example, a complete IDCT operation method is used as the IDCT operation method, and X
A partial IDCT operation method in which all the components other than the low frequency region represented by 0 to X3 are zero, and a case in which all the DC components other than the DC component in the DC region of x = y = 0 indicated by 1900 in FIG. 19 are zero. In the case of preparing four IDCT operation methods, a partial IDCT operation method and a partial IDCT operation method in which only one non-zero coefficient exists in an area other than the DC area (AC area), in an I picture or INTRA macro block, Since there is almost no block in which only one non-zero coefficient exists in the AC area, the partial IDCT operation method when only one non-zero coefficient exists in the AC area is excluded from the selection candidates of the IDCT operation method. be able to. Also, in consideration of high speed, if the image quality degradation allows image quality degradation of a B picture that does not propagate to other frames, the IDCT scheme selection unit for the B picture uses a complete IDCT method.
The operation method can be excluded from the selection candidates of the IDCT operation method. As described above, by having the IDCT scheme selection unit for each picture type or macroblock type, it is possible to efficiently select the IDCT calculation scheme in accordance with the characteristics of the distribution of the non-zero coefficients of each picture, and to effectively perform the computation. The amount can be reduced.

Further, assuming that all data up to the EOB code is non-zero data, it is not necessary to determine whether or not each coefficient is zero, so that it is possible to reduce the amount of calculation for detecting an effective coefficient.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION

[Explanation of Configuration] Next, an embodiment of the first invention will be described in detail with reference to FIG. Effective coefficient detector 100
Detects whether all four coefficients processed in parallel with DCT coefficients as input are zero and detects effective coefficients in block units. The switch 101 and the selector 105 switch according to the picture type. The effective coefficient detection result is transmitted to the first IDCT scheme selecting section 102 if the picture type is an I picture, to the second IDCT scheme selecting section 103 if the picture type is a P picture, or to the third IDCT scheme if the picture type is a B picture.
It is input to the method selection unit 104. The first IDCT scheme selection unit 102 selects a DC ID for an I-picture from a plurality of IDCT calculation schemes prepared in the IDCT calculation unit 107 in advance.
The candidates of the IDCT operation method are limited in consideration of the characteristics of the distribution of the T coefficient, and the IDCT operation method that minimizes the operation amount is selected from the candidates based on the effective coefficient detection result. The second IDCT scheme selection unit 103 preliminarily considers the characteristics of the distribution of DCT coefficients in a P picture from among a plurality of IDCT schemes prepared in the IDCT calculation unit 107, and limits IDCT calculation scheme candidates. I that minimizes the amount of calculation among the candidates based on the coefficient detection result
Select the DCT operation method. In the third IDCT scheme selection unit 104, a plurality of IDs prepared in the IDCT
Among the CT operation methods, IDCT operation method candidates are limited in advance in consideration of the distribution characteristic of DCT coefficients in a B picture, and the IDCT operation amount is the smallest among the candidates based on the effective coefficient detection result. Select a calculation method. I
The DCT operation unit 107 prepares a complete IDCT operation method for performing an IDCT operation without omitting all coefficients, and a plurality of partial IDCT operation methods for omitting an operation in consideration of a zero coefficient. Switch 106, selector 108
Switch based on the selection result of the IDCT scheme selection unit sent through the selector 105. IDCT operation unit 107
In the above, the IDCT operation is performed using the IDCT operation method selected by the IDCT method selection unit, and the operation result is output through the selector 108.

Next, an embodiment of the second invention will be described in detail with reference to FIG. The effective coefficient detection unit 200 detects whether all four coefficients processed in parallel with the DCT coefficient as input are zero, and detects an effective coefficient for each block. The switch 201 and the selector 204 switch according to the macroblock type. The effective coefficient detection result is output to the first IDCT scheme selecting section 202 if the macroblock type is an INTRA block, and to the second IDCT scheme selecting section 203 if the macroblock type is an INTER block. In the first IDCT scheme selection section 202, IDCT computation scheme candidates are limited in advance by considering the characteristics of the distribution of DCT coefficients in the INTRA block from among a plurality of IDCT computation schemes prepared in the IDCT computation section 206. Based on the coefficient detection result, an IDCT operation method that minimizes the operation amount is selected from the candidates. The second IDCT scheme selection section 203 limits the IDCT calculation scheme candidates in advance from among the plurality of IDCT calculation schemes prepared in the IDCT calculation section 206 in consideration of the distribution characteristic of DCT coefficients in the INTER block. Based on the coefficient detection result, an IDCT operation method that minimizes the operation amount is selected from the candidates. The IDCT calculation unit 206 includes a complete ID for performing IDCT calculation without omitting all coefficients.
There are prepared a CT operation method and a plurality of partial IDCT operation methods in which the operation is omitted in consideration of the zero coefficient. Switch 20
5. The selector 207 switches based on the selection result of the IDCT scheme selection unit sent through the selector 204. I
The DCT operation unit 206 performs the IDCT operation using the IDCT operation method selected by the IDCT method selection unit, and outputs the operation result through the selector 207.

Next, an embodiment of the third invention will be described in detail with reference to FIG. The effective coefficient detection unit 300 receives position information of an end-of-block (EOB) code, and detects an effective coefficient by regarding all DCT coefficients up to the position indicated by the EOB code as non-zero data. Switch 30
1, selector 305, first IDCT scheme selection unit 30
2, second IDCT scheme selection unit 303, third IDCT
A method selection unit 304, a switch 306, a selector 308,
The configuration of the IDCT operation unit 307 is the same as that of the first embodiment.

Next, an embodiment of the fourth invention will be described in detail with reference to FIG. The effective coefficient detection unit 400 receives position information of an end-of-block (EOB) code as input, regards all DCT coefficients up to the position indicated by the EOB code as non-zero data, and detects an effective coefficient. Switch 40
1, selector 404, first IDCT scheme selection unit 40
2. Second IDCT scheme selection unit 403, switch 40
5, the configuration of the selector 407 and the IDCT operation unit 406 are the same as those of the second embodiment.

Next, an embodiment of the fifth invention will be described in detail with reference to FIG. 8 points in the first invention ×
The eight-point two-dimensional IDCT device can be realized as a microprocessor program. It can be realized as a microprocessor program. The microprocessor 2201 is a microprocessor having a divided ALU operation for processing a plurality of operations with one instruction. The recording medium 2203 is the first
A program for realizing the processing performed by the 8 point × 8 point two-dimensional IDCT apparatus according to the invention of the present application is recorded. This recording medium 2203 may be a magnetic disk, a semiconductor memory, or another recording medium. The 8-point × 8-point two-dimensional IDCT program is transferred from the recording medium 2203 to the microprocessor 2201.
To control the operation of the microprocessor 2201. The microprocessor 2201 is an input device 220
Get picture type and DCT coefficient from 0. The DCT coefficient is temporarily written to the memory 2202. The microprocessor 2201 executes the same processing as the 8-point × 8-point two-dimensional IDCT device according to the first embodiment, and outputs the calculation result to the output device 2204.

Next, an embodiment of the sixth invention will be described in detail with reference to FIG. 8 points in the second invention ×
The eight-point two-dimensional IDCT device can be realized as a microprocessor program. The microprocessor 2301 is a microprocessor having a divided ALU operation for processing a plurality of operations with one instruction. The recording medium 2203 stores a program for realizing the processing performed by the 8-point × 8-point two-dimensional IDCT apparatus according to the second invention. This recording medium 2303 may be a magnetic disk, a semiconductor memory, or another recording medium. The 8-point × 8-point two-dimensional IDCT program is executed by the microprocessor 230
1 to control the operation of the microprocessor 2301. The microprocessor 2301 is the input device 23
From 00, the macroblock type and DCT coefficients are obtained.
The DCT coefficient is temporarily written to the memory 2302. The microprocessor 2301 is the same as the microprocessor 230 in the second embodiment.
The same processing as that of the point × 8 point two-dimensional IDCT device is executed, and the calculation result is output to the output device 2304.

Next, an embodiment of the seventh invention will be described in detail with reference to FIG. 8 points in the third invention ×
The eight-point two-dimensional IDCT device can be realized as a microprocessor program. The microprocessor 2401 is a microprocessor having a divided ALU operation for processing a plurality of operations with one instruction. The recording medium 2403 records a program for realizing the processing performed by the 8-point × 8-point two-dimensional IDCT apparatus according to the third invention. This recording medium 2403 may be a magnetic disk, a semiconductor memory, or another recording medium. The 8-point × 8-point two-dimensional IDCT program is transferred from the recording medium 2403 to the microprocessor 240
1 to control the operation of the microprocessor 2401. The microprocessor 2401 is the input device 24
From 00, the picture type, EOB address and DCT coefficient are obtained. The DCT coefficient is temporarily written to the memory 2402. The microprocessor 2401 executes the same processing as the 8-point × 8-point two-dimensional IDCT apparatus according to the third embodiment, and outputs the calculation result to the output device 2404.

Next, an embodiment of the eighth invention will be described in detail with reference to FIG. 8 points in the fourth invention ×
The eight-point two-dimensional IDCT device can be realized as a microprocessor program. The microprocessor 2501 is a microprocessor having a divided ALU operation for processing a plurality of operations with one instruction. The recording medium 2503 records a program for realizing the processing performed by the 8-point × 8-point two-dimensional IDCT apparatus according to the fourth invention. This recording medium 2503 may be a magnetic disk, a semiconductor memory, or another recording medium. The 8-point × 8-point two-dimensional IDCT program is transferred from the recording medium 2503 to the microprocessor 250.
1 to control the operation of the microprocessor 2501. The microprocessor 2501 is the input device 25
00 to macroblock type, EOB address and D
Obtain CT coefficients. The DCT coefficient is once written into the memory 2502. The microprocessor 2501 executes the same processing as the 8-point × 8-point two-dimensional IDCT device according to the fourth embodiment, and outputs the operation result to the output device 2504.

[Description of Operation] Next, the operation of the first invention will be described in detail. In the effective coefficient detection unit 100, D
It is checked whether all four coefficients processed in parallel with the CT coefficient as input are zero, and an effective coefficient is detected for each block. As shown in FIG. 5, the input DCT coefficients collectively represent X0 to X15 for every four coefficients processed in parallel.
The operation of the effective coefficient detection unit 100 will be described with reference to FIG.
First, in step 600, a flag (fbit) for storing a valid coefficient detection result and an index i are initialized. Next, at step 601, it is checked whether Xi is zero. If Xi is not zero, the i-th bit from the least significant bit of fbit is set to 1 in step 602, and the process proceeds to step 603. If Xi is zero, the process proceeds to step 603 without doing anything. In step 603, it is checked whether i = 15. If it is not i = 15, 1 is added to i in step 604, and the process returns to step 601. Steps 601 to 604 are repeated until i = 15. If i = 15 in step 603, fbit
Is output as a valid coefficient detection result, and the valid coefficient detection ends. As shown in FIG. 7, fbit representing the effective coefficient detection result indicates the state from X0 to X15 as valid (1) if there is a non-zero coefficient, and invalid (0) if it is zero. For example, if fbit is 0001 in hexadecimal, all bits except for X0 are zero.

The switching between the switch 101 and the selector 105 is performed in frame units, and after switching according to the picture type, it is fixed until the next picture. The effective coefficient detection result is the first if the picture type is an I picture.
Is output to the second IDCT scheme selection section 103 if the picture type is a P picture, and to the third IDCT scheme selection section 104 if the picture type is a B picture.

Next, the operation of the first, second, and third IDCT scheme selection units will be described with reference to FIG. IDCT operation unit 10
It is assumed that n (1 < n < N) IDCT operation methods have been previously selected as selection candidates from the N IDCT operation methods prepared in No. 7. First, at step 800, a counter j is initialized. In step 801, j and n are compared, and j <
If n, the process proceeds to step 802; otherwise, at step 804, the j (= n) -th IDCT operation method is selected. In step 802, it is checked whether the input effective coefficient detection result satisfies the condition for performing the n-th partial IDCT operation. If so, the j-th IDCT operation method is selected in step 804. If not, step 803 adds 1 to j, and the process returns to step 801.

The switch 106 and the selector 108 are IDC
Switching is performed according to the IDCT operation method selected by the T method selection unit. The IDCT operation unit 107 performs the IDCT operation using the IDCT operation method selected by the IDCT method selection unit, and outputs the operation result through the selector 108.

Next, the operation of the second invention will be described. The operation of the effective coefficient detector 200 is the same as that of the first invention. The switch 201 and the selector 204 switch on a macroblock basis according to the macroblock type. The macroblock type of the effective coefficient detection result fbit is INTR.
If it is an A block, it is output to the first IDCT scheme selection unit 202, and if the macroblock type is an INTER block, it is output to the second IDCT scheme selection unit 203. First IDCT scheme selecting section 202, second IDCT
A method selection unit 203, a switch 205, a selector 207,
The operation of the IDCT operation unit 206 is the same as that of the first invention.

Next, the operation of the third invention will be described. H. 2
61, significant D in image coding such as MPEG, JPEG, etc.
An end-of-block (EOB) code is added to the end of the CT coefficient, and data thereafter is omitted. The EOB code indicating the number of data after the EOB code
I will call it an address. The effective coefficient detection unit 300 detects an effective coefficient using the EOB address as an input. It is assumed that all coefficients up to the position indicated by the EOB address are non-zero data. When the DCT coefficient is sent in a zigzag scan as shown in FIG. 9B, the EOB address is 9 even when non-zero data exists only in the shaded area in FIG. 9A. In this case, it is assumed that all data up to the position indicated by the EOB address indicated by oblique lines in FIG. 9B are non-zero coefficients. Considering that four calculations are performed in parallel, only X0 to X3 shown in FIG. 9C are non-zero data. When the effective coefficient is detected in this way,
Effective coefficient detection result fbit when EOB address is 9
Can be expressed as 000f in hexadecimal. The result of the detection of the effective coefficient from the EOB address is output to the IDCT scheme selecting section as in the first invention. Switch and selector, ID
The operations of the CT method selection unit and the IDCT operation unit are the same as those of the first invention.

Next, the operation of the fourth invention will be described. The operation of the effective coefficient detection unit 400 is the same as the operation of the third invention, and includes a switch and a selector, an IDCT method selection unit,
The operation of the DCT operation unit is the same as that of the second invention.

[0032]

FIG. 10 shows an embodiment of the first invention. The effective area detection unit 1000 determines whether the DCT coefficient is zero according to the flowchart shown in FIG. 6 and outputs an effective area detection result fbit. fbit is a sum of four DCT coefficients processed in parallel as shown in FIG.
When expressed from 0 to X15, Xi (i = 0, 1, 2, 2,
, 15) are represented by 1/0 of the i-th bit counted from the least significant bit. Switching between the switch 1001 and the selector 1005 is performed on a picture basis according to the picture type. If the picture type is an I-picture, the result of the effective area detection is the first IDCT scheme selection unit 100
2 is output. If the logical product of fbit and hexadecimal number 0001 is 1, the first IDCT scheme selecting unit 1002 selects the partial IDCT operation scheme 1 and sets fbit and hexadecimal number 00.
When the logical product of 0f is 1, the partial IDCT operation method 2 is selected, and in other cases, the complete IDCT operation method is selected. When the picture type is a P picture, the effective area detection result is output to the second IDCT scheme selection unit 1003. The second IDCT scheme selection unit 1003 selects the partial IDCT scheme 3 if the logical product of fbit and the hexadecimal number 00ff is 1, and otherwise selects the complete IDCT scheme. When the picture type is a B picture, the effective area detection result is determined by the third IDCT scheme selection unit 100.
4 is output. The third IDCT scheme selection unit 1004 selects the partial IDCT scheme 1 if the logical product of fbit and hexadecimal 0001 is 1, and otherwise selects the partial IDCT scheme 2. IDCT operation unit 100
7 has four IDCT operation methods, a complete IDCT operation method and a partial IDCT operation method 1, 2, and 3. The switch 1006 and the selector 1008 are the selector 1
Switching is performed based on the selection result of the IDCT scheme selection unit transmitted through 005. The IDCT operation unit 1007 performs an IDCT operation using the selected IDCT operation method, and outputs an operation result through the selector 1008.

FIG. 15 shows an embodiment of the complete IDCT operation method. First, input DCT coefficients are transposed in order to simultaneously process four columns of 8 points × 8 points two-dimensional IDCT. R0, R
An 8-point one-dimensional IDCT indicated by 1 is executed. The operation results of R0 and R1 are transposed, and an 8-point one-dimensional IDC indicated by C0 and C1
T is executed, and the result is output as a result of the 8 points × 8 points two-dimensional IDCT. Any method may be used as the method of the eight-point one-dimensional IDCT operation.

FIG. 16 shows an embodiment of the partial IDCT operation method 1. In the partial IDCT operation method 1, the non-zero coefficient exists only in X0 of 1600, and the I
This is a DCT operation method. First, as R0, a one-dimensional IDCT operation in which operations other than X0 are omitted is performed. The calculation result of R0 is X0 'for all eight points. Since R1 has all zero input coefficients, the calculation is omitted. The operation result in the row direction is X in 1601
Only the data indicated by 0 'is non-zero, and all other data are zero. When the transposition in the xy direction is performed, the input data of the IDCT in the column direction is non-zero in only Y0 to Y3 of 1602 and all other data are zero, and the input data of C0 and C1 are both Y0.
To Y3. A one-dimensional IDCT operation in which operations other than Y0 to Y3 are omitted is performed in C0, and the operation result Y of C0 is performed.
From 0 'to Y7' are copied as the operation result of C1, and an 8-point × 8-point two-dimensional IDCT operation result 1603 is output.

FIG. 17 shows an embodiment of the partial IDCT operation method 2. The partial IDCT operation method 2 is an IDCT operation method in which non-zero coefficients are present only in X0 to X3 of 1700, and all other coefficients are zero. First, as R0, a one-dimensional IDCT operation in which operations other than X0 to X3 are omitted is performed. Since R1 has all zero input coefficients, the calculation is omitted. The operation results in the row direction are non-zero only in X0 'to X7' of 1701, and all other values are zero. When the operation result is transposed, the input data of the IDCT in the column direction is Y0 of 1702.
To Y3 and only Y8 to Y11 are non-zero coefficients, and all others are zero. In C0, a one-dimensional IDCT in which operations other than Y0 to Y3 are omitted is performed, and in C1, Y8 to Y9
8x8 by performing one-dimensional IDCT omitting operations other than
The calculation result 1703 of the point two-dimensional IDCT is output.

FIG. 18 shows an embodiment of the partial IDCT operation method 3. The partial IDCT operation method 3 is an IDCT operation method in which non-zero coefficients exist only in X0 to X7 of 1800 and all other coefficients are zero. First, an 8-point one-dimensional IDCT operation is performed without omission as R0. Since R1 has all zero input coefficients, the calculation is omitted. In the row-direction operation result, only X0 'to X7' of 1801 are non-zero, and all others are zero. The following operation is the partial IDCT operation method 2.
Is the same as

As a partial IDCT operation method, a method other than the above may be used. For example, when the DCT coefficient exists only in the DC component, the partial IDCT operation method 4 shown in FIG. 19 can be used. If only the DC component in the DC region of x = y = 0 indicated by the solid color at 1900 is a non-zero coefficient, the result of the 8-point × 8-point two-dimensional IDCT operation is 64
All points are the same. Therefore, a two-dimensional IDCT operation result a is calculated, and it is copied to four points as shown in 1901, and Y
Set to 0 '. Next, as indicated by 1902, Y0 'is copied and output as a result of an 8-point × 8-point two-dimensional IDCT.
When this method is used, the effective coefficient detector or IDC
It is necessary to check that only the DC component has a zero coefficient in the T-system selector.

When only one DCT coefficient exists in the AC area, the partial IDCT operation method 5 shown in FIG. 20 can be used. DCT coefficient f (x, y) in AC domain
(X, y = 0, 1, 2,... 7 except for x = y = 0), there is only one non-zero coefficient A, and the other coefficients are all zero. 8 points x 8 points 2D ID
The CT operation result T (x, y) is calculated in advance and stored in a table. If only one non-zero coefficient f (x, y) = a exists in the AC domain as shown in 2000 and the other DCT coefficients are zero, 64 elements of the table T (x, y) 8 × 8 points 2 by multiplying each by a / A
A dimension IDCT operation result 2001 can be calculated. The multiplication multiplied by a / A can also be performed in four parallel. When this method is used, the effective coefficient detecting unit or the IDCT method selecting unit detects the position information x, y of only one non-zero coefficient,
It is necessary to send to the IDCT operation part.

The judgment method of the IDCT method selection section changes according to the prepared partial IDCT operation method.

FIG. 11 shows an embodiment of the second invention. The effective area determination unit 1100 determines whether the DCT coefficient is zero according to the flowchart shown in FIG. 6 and outputs an effective area detection result fbit. Switch 1101,
The selector 1104 switches for each macroblock according to the macroblock type. Macro block type is I
In the case of NTRA Block, the effective area detection result is output to IDCT first scheme selection section 1102. If the logical product of fbit and hexadecimal number 0001 is 1, the first IDCT method selecting unit 1102 selects the partial IDCT operation method 1, and the logical product of fbit and hexadecimal number 000f is 1
If so, the partial IDCT operation method 2 is selected, and in other cases, the complete IDCT operation method is selected. When the macro block type is the INTER block, the effective area detection result is output to the second IDCT scheme selection unit 1103. When the logical product of fbit and hexadecimal number 000f is 1, the second IDCT scheme selecting unit 1103 sets the partial IDCT
The operation method 2 is selected, and in other cases, the complete IDCT operation method is selected. The IDCT calculation unit 1106 has a complete IDC
T operation method and partial IDCT operation methods 1 and 2 are prepared. The switch 1105 and the selector 1107 switch according to the selection result of the IDCT scheme selection unit output through the selector 1104. The IDCT operation unit 1106 performs the IDCT operation using the selected IDCT operation method, and outputs the operation result through the selector 1107.

FIG. 12 shows an embodiment of the third invention. Effective coefficient detection section 1200 calculates fbit from the EOB address. Assuming that all the DCT coefficients up to the EOB address are non-zero coefficients, for example, when the EOB address is 9, all coefficients at positions represented by numbers from 0 to 9 in 9.2 are considered to be non-zero. Will be. Considering the calculation in four parallels, X0, X1, X2, and X3 are non-zero, and x4 to x15 are all zero. Therefore, fbit can be represented by hexadecimal 000f. Thus, fbit is calculated according to the EOB address. I
The same DCT scheme selection unit, IDCT calculation unit, switch, and selector as those in the first embodiment can be used.

FIG. 13 shows an embodiment of the fourth invention. As the effective coefficient detecting unit 1300, the one shown in the embodiment of the third invention is used, and the other parts can be the same as those of the second embodiment.

The effective coefficient detector in the third and fourth aspects of the present invention does not calculate fbit from the EOB address.
The B address itself can be used as the effective coefficient detection result. In this case, if the EOB address is 0, the IDCT method selection unit selects the partial IDCT operation method 1
If the B address is 9 or less, the partial IDCT operation method 2 is selected, otherwise the complete IDCT operation method is used.

The eight-point by eight-point two-dimensional I shown in FIGS. 10 to 13 according to the first to fourth embodiments of the present invention as described above.
The DCT device can be realized as a program of a microprocessor that executes the same process for a plurality of instructions with one instruction.

[0045]

According to the first aspect of the present invention, the amount of calculation can be effectively reduced according to the characteristics of the distribution of DCT coefficients for each picture type. The reason is that the selection candidates of the IDCT operation method are limited according to the characteristics of the distribution of the DCT coefficient for each picture type, so that the partial IDCT operation is selected in many coding blocks while suppressing the operation related to the IDCT method selection. This is because the amount of calculation can be reduced.

According to the second aspect of the present invention, the amount of calculation can be effectively reduced according to the characteristics of the distribution of DCT coefficients for each macroblock type. The reason is that the selection candidates of the IDCT operation method are limited in accordance with the characteristics of the distribution of the DCT coefficients for each macroblock type.
This is because the amount of calculation can be reduced by selecting the T calculation.

According to the third and fourth aspects of the present invention, the amount of calculation of the effective area detecting section can be reduced. The reason is that, instead of determining whether or not all the coefficients are zero, the effective coefficient determination is performed on the assumption that all the coefficients up to the EOB address are non-zero coefficients.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first invention.

FIG. 2 is a block diagram showing a configuration of the second invention.

FIG. 3 is a block diagram showing a configuration of the third invention.

FIG. 4 is a block diagram showing a configuration of the fourth invention.

FIG. 5 is an explanatory diagram showing input DCT coefficients.

FIG. 6 is a diagram for explaining the operation of an effective area detection unit.

FIG. 7 is an explanatory diagram showing an effective area detection result fbit.

FIG. 8 is a diagram for explaining the operation of an IDCT scheme selection unit.

FIG. 9 is a diagram for explaining the operation of an effective area detection unit using an EOB address.

FIG. 10 is a block diagram showing an embodiment of the first invention.

FIG. 11 is a block diagram showing an embodiment of the second invention.

FIG. 12 is a block diagram showing an embodiment of the third invention.

FIG. 13 is a block diagram showing an embodiment of the fourth invention.

FIG. 14 is an explanatory diagram showing a Row-Column method.

FIG. 15 is an explanatory diagram showing an embodiment of a complete IDCT operation method.

FIG. 16 is an explanatory diagram showing an embodiment of a partial IDCT operation method 1;

FIG. 17 is an explanatory view showing an embodiment of a partial IDCT operation method 2;

FIG. 18 is an explanatory view showing an embodiment of a partial IDCT operation method 3;

FIG. 19 is an explanatory diagram showing an embodiment of a partial IDCT operation method 4.

FIG. 20 is an explanatory diagram showing an embodiment of a partial IDCT operation method 5.

FIG. 21 is a block diagram showing a configuration of a conventional example.

FIG. 22 is a block diagram showing a configuration of the fifth invention.

FIG. 23 is a block diagram showing a configuration of the sixth invention.

FIG. 24 is a block diagram showing a configuration of the seventh invention.

FIG. 25 is a block diagram showing a configuration of the eighth invention.

[Explanation of symbols]

100, 200, 300, 400, 1000, 110
0, 1200, 1300, 2100 Effective coefficient detector 101, 106, 201, 205, 301, 306, 4
01,405,1001,1006,1101,110
5,1201,1206,1301,1305,210
2 switches 102, 103, 104, 202, 203, 303, 3
04, 402, 403, 1002, 1003, 100
4,1102,1103,1202,1203,120
4,1302,1303,2101 IDCT scheme selection unit 107,206,307,406,1007,110
6, 1207, 1306, 2103 IDCT operation unit 105, 108, 204, 207, 305, 308, 4
04,407,1005,1008,1104,110
7, 1205, 1208, 1304, 1307, 210
4 Selector 2200, 2300 Input device 2201, 2301, 2401, 2501 Microprocessor 2202, 2302, 2402, 2502 Memory 2203, 2303, 2403, 2503 Recording medium 2204, 2304, 2404, 2504 Output device

Claims (8)

(57) [Claims] 1. An 8-point × 8-point 8-point one-dimensional inverse discrete cosine transform (IDCT) is executed in parallel by using a processor that executes the same processing on a plurality of data with one instruction.
In the point two-dimensional IDCT apparatus, an effective coefficient detection unit that determines whether four DCT coefficients to be simultaneously processed are non-zero, and an IDCT operation method that requires the least amount of operation based on the effective coefficient detection result. 1st, 2nd, 3rd I
A DCT scheme selection section, a first switch and a first selector for switching the IDCT scheme selection section according to picture type, an IDCT computation section having a plurality of IDCT computation schemes, and an IDCT computation selected by the IDCT scheme selection section A second switch and a second selector for switching an IDCT operation method used in the IDCT operation unit according to a system, wherein the IDCT operation unit performs an I type for performing intra-frame encoding and performs forward prediction encoding Depending on the characteristics of the distribution of the DCT coefficients in each of the P-type and B-type picture types that perform bidirectional predictive encoding, some IDs are set in advance.
A CT operation method is prepared, and the first switch and the first selector are switched according to a picture type. If the picture type is an I picture, the first I
DCT method selection unit, if P picture, the second IDC
In the case of a T picture, if the picture is a B picture, the third IDCT method selecting section selects the IDCT calculation method capable of reducing the amount of calculation in accordance with the effective coefficient detection result.
An eight-point by eight-point two-dimensional IDCT apparatus, wherein the second switch and the second selector are switched according to a DCT operation method, and an IDCT operation is performed according to the IDCT operation method selected by the IDCT method selector. . 2. An 8-point × 8-point 8-point one-dimensional inverse discrete cosine transform (IDCT) executed in parallel by using a processor that executes the same processing on a plurality of data with one instruction.
In the point two-dimensional IDCT apparatus, an effective coefficient detection unit that determines whether four DCT coefficients to be simultaneously processed are non-zero, and an IDCT operation method that requires the least amount of operation based on the effective coefficient detection result. 1, the second IDCT
A method selection unit, a first switch and a first selector for switching the IDCT method selection unit according to the macroblock type, an IDCT operation unit having a plurality of IDCT operation methods, and an IDCT operation selected by the IDCT method selection unit A second switch and a second selector for switching an IDCT operation method used in the IDCT operation unit in accordance with a system; an INTRA type for performing intra-frame encoding; an IN for performing inter-frame encoding;
DCT in each macroblock type of TER type
Several IDCT calculation methods are prepared in advance according to the characteristics of the coefficient distribution, and the first switch and the first selector are switched according to the macroblock type. 1
IDCT scheme selection unit, if INTER, the second I
The DCT method selection unit selects the IDCT operation method capable of reducing the amount of operation most according to the effective coefficient detection result, and switches the second switch and the second selector according to the IDCT operation method to switch the IDCT operation method. An eight-point by eight-point two-dimensional IDCT apparatus that performs an IDCT operation according to an IDCT operation method selected by a method selection unit. 3. An 8 point × 8 point two-dimensional IDC according to claim 1.
In the T device, an effective coefficient detection unit receives position information of an end-of-block (EOB) code as input, and performs effective coefficient detection by regarding all DCT coefficients up to the position indicated by the EOB code as non-zero coefficients. 8 points × 8 points two-dimensional IDCT apparatus. 4. The 8-point × 8-point two-dimensional IDC according to claim 2.
In the T device, an effective coefficient detection unit receives position information of an end-of-block (EOB) code as input, and performs effective coefficient detection by regarding all DCT coefficients up to the position indicated by the EOB code as non-zero coefficients. 8 points × 8 points two-dimensional IDCT apparatus. 5. A process for writing an input DCT coefficient to a memory and a process for performing the process by an 8 point × 8 point two-dimensional IDCT apparatus according to claim 1, wherein the same process is performed for a plurality of data by one instruction. A recording medium on which a program to be executed by a microprocessor to be executed is recorded. 6. A process for writing an input DCT coefficient into a memory and a process for performing the process by an 8 point × 8 point two-dimensional IDCT apparatus according to claim 2 by executing the same process with a single instruction for a plurality of data. A recording medium on which a program to be executed by a microprocessor to be executed is recorded. 7. A process of writing an input DCT coefficient to a memory and a process of executing an 8 point × 8 point two-dimensional IDCT apparatus according to claim 3 by executing the same process with a single instruction for a plurality of data. A recording medium on which a program to be executed by a microprocessor to be executed is recorded. 8. A process for writing the input DCT coefficient into a memory and a process for performing the process by the 8-point × 8-point two-dimensional IDCT apparatus according to claim 4, wherein the same process is executed for a plurality of data by one instruction. A recording medium on which a program to be executed by a microprocessor to be executed is recorded.