JP2639842B2 - Discrete cosine transform coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] Discrete cosine transform coding, which is one of high-efficiency coding schemes, is intended to improve the coding efficiency by limiting the allocated codes in a high-frequency domain. Unit, a quantization unit, an encoding unit, and a maximum value detection unit, and performs a discrete cosine transform in the discrete cosine transform unit in correspondence with a block including a plurality of pixels of the input image signal, and performs the quantization. Quantizing the transform coefficient of the discrete cosine transform output in the section, detecting the maximum value of the quantized output signal in the low frequency region by the maximum value detection section,
A value obtained by adding a constant to the absolute value of the maximum value is set as a limit value on the high frequency region side, and codes up to the limit value are assigned and encoded by the encoding unit.

[Industrial applications]

The present invention relates to a discrete cosine transform coding method, which is one of high efficiency coding methods.

In videoconferencing, videophone, etc., various high-efficiency coding schemes for performing band compression of image signals have been proposed.
Recently, discrete cosine transform (Discrete C
osine transform) is attracting attention. In the discrete cosine transform, a plurality of pixels are treated as one block, and the transform coefficients are orthogonally transformed by a transform matrix. The transform coefficients generally concentrate on the low frequency region side, although they vary depending on the properties of the image. Then, the transform coefficients are entropy-encoded and transmitted. On the receiving side, entropy decoding and inverse quantization are performed, and then inverse discrete cosine transform is performed to obtain a reproduced image signal.

In such a discrete cosine transform coding method, there is a demand for further improving the efficiency.

[Conventional technology]

FIG. 23 is a block diagram of a conventional example, in which a transmitting unit includes a discrete cosine transform unit 30, a quantizing unit 31, a scanning unit 32, and an encoding unit 33.
The receiving unit includes a decoding unit 34, an inverse scanning unit 35, an inverse quantization unit 36, and an inverse discrete cosine transform unit 37.

The input image signal is a signal obtained by digitizing an image signal from a television camera or the like or a signal obtained by performing an inter-frame difference process on the image signal, and one of the input image signals.
For example, as shown in FIG. 24, the screen portion is divided into blocks of N × N pixels, and the pixels f (u, v) of each block are divided.
Is subjected to a discrete cosine transform in a discrete cosine transform unit 30, and the transform coefficients F (i, j) are individually quantized in a quantizing unit 31. In a scanning unit 32, horizontal, vertical or zigzag directions are obtained. Is performed, the quantized output signals in the two-dimensional array are converted into a one-dimensional array, and the encoding unit 33 sequentially performs entropy encoding and transmits the entropy-encoded signals to the receiving side.

On the receiving side, entropy decoding is performed in the decoding unit 34, processing reverse to that of the scanning unit 32 on the transmitting side is performed in the reverse scanning unit 35, and the quantization unit 31 is processed in the inverse quantization unit 36. The inverse discrete cosine transform unit 37 performs inverse quantization of the process, and the inverse cosine transform unit 37 performs the inverse process of the discrete cosine transform unit 30 to reproduce an image signal, which is added to a display device (not shown).

The discrete cosine transform in the discrete cosine transform unit 30 is
It is expressed by the following equations (1) and (2).

Using the equations (1) and (2), the pixel f
When (u, v) is converted to F (i, j), the conversion coefficient F (i, j)
j) indicates a frequency component corresponding to the value of i, j. The smaller the value of i, j, the lower the frequency component, and F (0,0)
Indicates a DC component. For example, assuming that N = 8 in the block of N × N pixels in the screen of FIG. 24, the shaded block is shown enlarged below and is subjected to discrete cosine transform. The direction of arrow a is a conversion coefficient in a low frequency region, and the direction of arrow b is a conversion coefficient in a high frequency region. Then, in the case of a general screen, the value of the low frequency component is large by performing the discrete cosine transform,
The value of the high-frequency component is small, and the value of the component that becomes 0 increases.

The conversion coefficient F (i, j) corresponding to the block is quantized in the quantization unit 31 and scanned in the scanning unit 32.
In the lower block of FIG. 24, c is a zigzag scan,
d indicates horizontal scanning, and e (dot-dash line) indicates vertical scanning. In any one of these scanning directions, scanning is performed from a low-frequency area to a high-frequency area or in the opposite direction, and a two-dimensional array is formed. Into a one-dimensional array, and the encoding unit 33
, The entropy coding is performed sequentially.

As described above, since the high-frequency component is often 0, scanning is performed from the high-frequency region side, and an end-of-bound code EOB (End Of Bound) is placed at a position where it is no longer 0.
Is added. For example, 8 × 8 pixels (1) shown in FIG.
The quantized signal of the block (64) is subjected to zigzag scanning from the high-frequency region side, and an end-of-bound code EOB is added to a position that is not zero (a position of a value of -7; a position of a pixel number (46)). . Thus, when encoding and transmitting from the low frequency domain side, the end-of-bound code EO
Since the high-frequency components of the pixel numbers (47) to (64) after the position B are all 0, it is not necessary to transmit, and the encoding efficiency can be improved.

On the receiving side, it is possible to identify the end of the coded signal of the block by using the end-of-bound code EOB and perform the decoding process.

FIG. 26 and FIG. 27 are explanatory diagrams of the code table of the conventional example, and show the range of −15 to +15 of the output of the quantization unit.
FIG. 26 shows a case of encoding with a 5-bit fixed code including the end-of-bound code EOB, and FIG. 27 shows a case of encoding with a 4- to 6-bit variable length code. FIG. 28 shows a case where the code table of FIG. 26 is expanded into a tree, and FIG. 29 shows a case where the code table of FIG. 27 is expanded into a tree. In FIGS. 27 and 29, the entire information amount is reduced by assigning a short code to a value near 0 having a relatively high occurrence probability.

FIG. 30 shows the distribution of frequency components by the discrete cosine transform. The same numerical value is the same frequency component group, and the larger the numerical value is, the higher the frequency region is. Therefore, from the low frequency region to the high frequency region,
Alternatively, when zigzag scanning is performed from the high frequency region to the low frequency region, a one-dimensional array in which the frequency component groups are sequentially arranged is obtained.

[Problems to be solved by the invention]

By performing discrete cosine transform on the block corresponding to the input image signal, transform coefficients concentrated on the low frequency region side can be obtained, and the value tends to be smaller in the high frequency region than in the low frequency region. It is.

However, conventionally, even if entropy coding is performed, the same code is assigned to both the low-frequency region and the high-frequency region, so that a codeword is assigned to a value that is unlikely to occur in the high-frequency region. Thus, the coding efficiency cannot be sufficiently improved.

An object of the present invention is to improve the coding efficiency by limiting the allocated codes in the high frequency region.

[Means for solving the problem]

The discrete cosine transform coding method of the present invention is based on the premise that the value on the low frequency region side is large and the value on the high frequency region side is small. The encoding is performed by restricting the value to the value obtained by adding.

Discrete cosine transform unit 1, quantization unit 2, encoding unit 3
And a maximum value detection unit 4. The discrete cosine transformation unit 1 performs a discrete cosine transformation on a block composed of a plurality of pixels of the input image signal, and a quantization unit 2 converts the discrete cosine transformation output. The coefficient is quantized, and the maximum value detection unit 4
Thus, the maximum value of the quantized output signal in the low frequency region is detected, and a value obtained by adding a constant to the absolute value of the maximum value is set as a limit value in the high frequency region. Is encoded.

[Action]

The discrete cosine transform unit 1 performs discrete cosine transform on a block-by-block basis using N × N pixels of an input image signal, and the quantizer 2 quantizes transform coefficients of a discrete cosine transform output. It is. Maximum value detector 4
Detects the maximum value of the quantized output signal on the low frequency side,
This limits the maximum value on the high frequency region side. For example,
When detects a maximum value M ₃ in column "3" in the low frequency range side of Figure 30, the maximum value the absolute value the values of more than a value obtained by adding a constant α in of M ₃ is "3" It is assumed that no occurrence occurs in the row of “4” adjacent to the high-frequency area side of the row, and (M ₃ + α)
Even if a value equal to or more than the value of “4” occurs in the column “4”, the value is limited to that value. Accordingly, it is possible to set a shorter code to be assigned to the high frequency region side, and it is possible to improve coding efficiency.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 2 is a block diagram of an embodiment of the present invention. The transmitting unit includes a discrete cosine transform unit 11, a quantizing unit 12, an encoding unit 13 for performing entropy encoding, a maximum value detecting unit 14, A scanning unit 15 and a limiter 16 are provided, and the receiving unit includes a decoding unit 16 that performs entropy decoding, an inverse scanning unit 17, a maximum value detection unit 18, an inverse quantization unit 19, and an inverse discrete cosine transform. And a unit 20.

The discrete cosine transform unit 11 subjects the input image signal to N × N pixels as one block, performs discrete cosine transform for each block, and adds the transform coefficient to the quantization unit 12 to quantize. The scanning unit 15 converts the quantized output signal of the two-dimensional array corresponding to the block into a one-dimensional array, and adds it to the limiter 16 and the maximum value detecting unit 14.

The maximum value detection unit 14 detects, for example, the maximum value in a row of diagonal scanning in the case of zigzag scanning,
The value obtained by adding the constant α to the absolute value of the maximum value is added to the limiter 16 and the encoding unit 13 as a limit value. In the limiter 16, even if the maximum value of the next row on the high frequency region side is equal to or larger than the limit value, the maximum value is limited according to the limit value. Further, the encoding unit 13 performs entropy encoding by controlling the number of code allocation bits.

In the receiving unit, the decoding unit 16 performs entropy decoding, the inverse scanning unit 17 converts the one-dimensional array into a two-dimensional array corresponding to the block, and the maximum value detecting unit 18 outputs, for example, each column of the frequency component group. , And controls the decoding unit 16. Further, the decoded signal converted into a two-dimensional array by the inverse scanning unit 17 is inversely quantized by the inverse quantization unit 19, and inverse discrete cosine transform is performed by the inverse discrete cosine transform unit 20 to obtain an output image signal. The received image is displayed by adding it to a display device (not shown) or the like.

FIG. 3 is a flowchart showing the operation of the maximum value detection unit 14. One block is set to 8 × 8, the scanning unit 15 converts the quantized output signal into a one-dimensional array into S (8 × 8), and sets a constant to α. , End of column by zigzag scanning of two-dimensional array
(13) and (a), the end of which is shown as the pixel number in FIG. The first column of the two-dimensionally arrayed quantized output signal indicates a DC component. Therefore, the maximum value is not detected for K = 1 of S (K).
(B), and after setting the maximum detection value MAX2 = 0 (c), S
It is determined whether or not the absolute value of (K) is larger than the maximum value MAX2 (d). If not, K = K + 1 is set (f). If it is larger, the absolute value of S (K) is set as the detection maximum value MAX2. (E), K = K + 1 (f), it is determined whether or not K> 3 (g). If not K> 3, the process proceeds to step (c), and
If> 3, I = 1 (h) and the process proceeds to the next step. That is, the columns I and 1 for which the maximum value has been detected are set to 1, and the maximum value detection for the first column of the 13 columns of the zigzag scan excluding the first and the last of the 8 × 8 two-dimensional array ends.

Then, the absolute value obtained by adding the constant α to the maximum value MAX2 obtained in step (e) is set as the maximum value MAX1 of the next column (i), and is notified to the encoding unit 13 that performs the entropy encoding.

Next, MAX2 = 0 is set (j), and S (K) = MIN (MAX (S
(K), -MAX1), MAX1) (k). That is, ± MAX1
Is limited to the range of the quantized output signal S (K).

Next, similarly to step (d), it is determined whether or not the absolute value of S (K) is larger than the maximum detection value MAX2 (l). If not, K = K + 1 (n). S
The absolute value of (K) is defined as the maximum detection value MAX2 (m), and K = K +
1 (n).

Next, it is determined whether or not K> M (I) (o). That is, it is determined whether or not it is the end of the column. If it is not the end, the process proceeds to step (k). If it is the end, I = I + 1 (p), and it is determined whether I> 13 (q), If it is not I> 13, the process goes to step (i) to detect the maximum value for the next column, and if I> 13, the maximum value detection process for one block is ended.

In addition, horizontal scanning or vertical scanning may be performed to detect the maximum value of each column, and a process of limiting the maximum value of the next column may be performed. The constant α can be set to a different value for each column of scanning.

FIG. 4 is a flowchart of one embodiment of the present invention,
A code table T, a provisional table T ', and a constant A (4) = (7,3,1,0) for the quantized output signal of -15 to +15 are determined (1), and the input from the maximum detector, that is, When the maximum value MAX is added (2), the maximum value MAX and a constant A (1) =
7, A (2) = 3, A (3) = 1, A (4) = 0 are compared with (3), (4), (5), (6), and the maximum value MAX is 7
If not less than 3, if not less than 1, if not less than 1, if not less than 0, process when less than 0,
That is, the process of creating the temporary table T 'is performed. Note that the code table T can be the same as the configuration shown in FIG.

FIG. 5 shows a flowchart of the process, where I = −15.
(A1), then set T '(I) = T (I) (a2), then set I = I + 1 (a3), and determine whether I> 15 (a4)
If not I> 15, repeat the above steps, I> 15
Then, the same temporary table T 'as the code table T is formed, and encoding is performed using this temporary table T' (a5).

FIG. 6 shows a processing flowchart of FIG.
= −15 (b1), and it is determined whether I> 0 and I / 2 = 0 (b
2) In the case of the condition, the EOB of the code table T excluding the rightmost bit is set as the EOB of the temporary table T ′ (b
3) If the condition is not satisfied, α ′ is obtained by removing the rightmost bit of the code table T, and the α ′ is made to correspond to the value of I / 2 of the temporary table T ′ (b4). And I = I +
It is set to 1 (b5), it is determined whether or not I> 15 (b6), and the process is repeated until I> 15 to form a temporary table T ', and encoding is performed using the temporary table T' (b7). In this case, a provisional table T ′ having a 4-bit structure is provided for a code table T having a 5-bit structure.

FIGS. 7, 8, and 9 show flowcharts of the processing of (c1) to (c1), respectively.
7), (d1) to (d7), (e1) to (e7) correspond to steps (b1) to (b7) in FIG. 6, and basically, FIG. A temporary table T 'having a 3-bit structure is formed by removing the right two bits of T. FIG. 8 forms a temporary table T' having a 2-bit structure by removing the right three bits of the code table T. FIG. A temporary table T 'having a 1-bit configuration is formed by removing the right 4 bits of the code table T. Also, except for FIGS. 7, 8, and 9, I / 4, I / 8, I /
When 16 becomes 0, if I> 0, it is EOB and I <0
Then, it is set to 0. Then, encoding is performed using the temporary table T '.

FIG. 10 to FIG. 13 are explanatory diagrams of a tree structure of a code,
FIG. 10 shows the case of the temporary table T 'having a 4-bit structure formed by the processing shown in FIG. 6, and FIG. 11 shows the case of the temporary table T' having a 3-bit structure formed by the processing shown in FIG.
FIG. 12 shows the case of a temporary table T 'having a 2-bit structure formed by the processing shown in FIG.
The figure shows the case of a temporary table T 'having a 1-bit structure formed by the processing shown in FIG.

FIG. 14 is an explanatory view of the temporary table, and shows temporary tables Ta1 to Ta4 formed by the processing of FIGS. 6 to 9 based on the code table shown in FIG. That is, the code table having a 5-bit configuration of the code table for coding the quantized output signal of −15 to +15 is deleted from the temporary table Ta1 by one bit to the right.
A 4-bit codeword for encoding the quantized output signal of 7 to +7 is used. The temporary table Ta2 is deleted by 2 bits on the right side and has a 3-bit configuration for encoding the quantized output signal of -3 to +3. The temporary table Ta3 is a 2-bit codeword for encoding the quantized output signal of −1 to +1 by deleting the right three bits, and the temporary table Ta4 is deleted by the right four bits to be 0. This is a one-bit codeword for encoding EOB. EOB is provided between “0” and “1”.

5 to 9 based on the code table shown in FIG.
A temporary table can be formed by a process similar to the process shown in the figure. FIG. 15 to FIG. 18 are explanatory diagrams of the tree structure of the code in that case. For example, in FIG.
A codeword of 5 bits to 3 bits is assigned to the quantized output signal of -7 to +7. EOB is provided between “0” and “1”.

FIG. 19 is an explanatory diagram of the temporary table, which is shown in FIG.
Based on a code table in which code words of 6 to 4 bits are allocated to the quantized output signals of 15 to +15, the temporary table Tb1 is
Bits are deleted to assign a code word of 5 to 3 bits to the quantized output signal of -7 to +7.
Delete the right 2 bits and add 4 to -3 to +3 quantized output signal.
A configuration in which a 2-bit code word is assigned, and the provisional table Tb3
Encodes the quantized output signal of -1 to +1 by deleting the right three bits. However, since it can be encoded by two bits including the EOB, the right four bits are deleted as a result. Corresponds to the case. Further, the temporary table Tb4 is for deleting 0 bits on the right and encoding 0 and EOB with 1 bit.
As a result, this corresponds to the case where the right 5 bits are deleted.

FIG. 20 is a flow chart of still another embodiment of the present invention, showing a case where the temporary table T 'of the above-described embodiment is formed in advance, and code tables T1 to T7 are selected according to the detected maximum value. As a code table, a code table T1 including a quantized output signal of −15 to 15 and EOB, −10 to +10, and E
A code table T2 including OB, a code table T3 including −7 to +7 and EOB, a code table T4 including −5 to +5 and EOB, and a code table T5 including −3 to +3 and EOB.
And a code table T6 including −1 to +1 and EOB, and 0 and E
A code table T7 for the OB is determined (11).

The maximum value MAX (12) of the input from the maximum value detector and the constant 10,
Compare with 7,5,3,1,0 respectively (12), (13), (14),
(15), (16), (17), (18), maximum value MAX is constant 10
In the above case, the code table T1, when the value is smaller than the constant 10 and 7 or more, the code table T2. When the value is smaller than the constant 7 and 5 or more, the code table T3. Code table T when small and not 1
5. The code table T6 when the value is smaller than the constant 3 and not 0, and the code table T7 when the value is 1 or 0 smaller than the constant 3 is selected and encoded.

FIGS. 21 and 23 show code tables T2 to T7 when the code table T1 is the code table of the variable length code shown in FIG. 27, and each of the code tables T2 to T7 is the same as that of the above-described embodiment. Is created in advance from the code table T1 in accordance with the procedure for creating the temporary table T '. In each of the code tables T2 to T7 in this case, between "0" and "1"
EOB is provided.

Then, the quantized output signal is encoded using the code table selected according to the maximum value MAX from the maximum value detection unit, and the code word in the high frequency region can be shortened. Efficiency can be improved.

The constant α to be added to the maximum value detected by the maximum detection unit 14 can be selected according to the nature of the image, and is, for example, about 1% to 10% of the value of the maximum quantization output. be able to. It is also possible to sequentially change the constant α to be smaller for each scanning row. Also, the present invention
It is also possible to apply in combination with other means for improving coding efficiency. Further, the configuration of each unit such as the discrete cosine transform unit 11 and the maximum value detecting unit 14 can be easily realized by using an arithmetic function such as a microprocessor.

〔The invention's effect〕

As described above, according to the present invention, the discrete cosine transform unit 1 performs discrete cosine transform of an input image signal in block correspondence, quantizes the quantized data in the quantizing unit 2, and outputs the maximum value in the scanning row by the maximum value detecting unit 4. , And a value obtained by adding a constant α to the absolute value of the maximum value is set as a limit value of the scan line on the high frequency region side from the scan line, and a code up to the limit value is assigned to the encoding unit 3. By performing the discrete cosine transform, the value on the high frequency region side tends to be reduced, and even if a large value occurs on the high frequency region side, the constant α is set to the maximum value on the low frequency region side. Are sequentially set and codewords greater than the limit value are not assigned, so that there is an advantage that the total information amount can be reduced.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the principle of the present invention, FIG. 2 is a block diagram of an embodiment of the present invention, FIG. 3 is a flowchart of a maximum value detecting section of one embodiment of the present invention, and FIGS. FIG. 10 to FIG. 13 are explanatory diagrams of a symbol tree structure, FIG. 14 is an explanatory diagram of a temporary table, FIGS. 15 to 18 are explanatory diagrams of a symbol tree structure, 19 is an explanatory diagram of a temporary table according to another embodiment of the present invention, FIG. 20 is a flowchart of still another embodiment of the present invention, FIGS. 21 and 22 are explanatory diagrams of a code table, and FIG. Block diagram of conventional example,
FIG. 24 is a diagram for explaining scanning of blocks, FIG. 25 is a diagram for explaining scanning of a conventional example, FIGS. 26 and 27 are diagrams for explaining a code table of a conventional example, and FIGS. 28 and 29 are diagrams for explaining a tree structure of codes. Fig. 30
The figure is an explanatory diagram of frequency components. 1 is a discrete cosine transform unit, 2 is a quantization unit, 3 is an encoding unit, and 4 is a maximum value detection unit.

Claims (1)

(57) [Claims] A block comprising a plurality of pixels of an input image signal, comprising a discrete cosine transform unit (1), a quantization unit (2), an encoding unit (3), and a maximum value detection unit (4). Correspondingly, a discrete cosine transform is performed in the discrete cosine transform unit (1), a transform coefficient of the discrete cosine transform output is quantized in the quantizing unit (2), and the maximum value detecting unit (4) The maximum value of the quantized output signal in the low frequency region is detected, a value obtained by adding a constant to the absolute value of the maximum value is set as a limit value in the high frequency region, and a code up to the limit value is assigned. A discrete cosine transform coding method characterized by coding in (3).