JPH08180194A - Method and device for coding - Google Patents

Abstract

PURPOSE: To converte the orthogonal transformation coefficient data into the orthogonal transformation coefficient data of a 1/n reduced image without an orthogonal inverse transformation and conversion of the resolution in an actual space. CONSTITUTION: In an input terminal 10, JPEG coded image data is inputted. The data is returned to DCT conversion coefficient data by a variable length decoding circuit 12 and an inverse quantization circuit 14 and the data is temporarily stored. A conversion coefficient synthesizing circuit 20 synthesizes 4 longitudinally and laterally adjacent conversion coefficient blocks which are stored in a memory 16 into a conversion coefficient block by a matrix calculation. The synthetic result of the circuit 20 is quantized by a quantization circuit 22 and is coded by a variable length coding circuit 24.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a coding apparatus and method, and more particularly to a coding apparatus and method for converting image code data that has been orthogonally transformed and coded into image code data of a reduced image. .

[0002]

2. Description of the Related Art When it is desired to reduce an image or change a sample ratio, original image data, if any, may be subsampled or filtered and then subsampled. On the other hand, when only image data that has already been subjected to orthogonal transform coding can be obtained, it is necessary to perform orthogonal inverse transform once to return it to the spatial data of the original image and subsample it. Since the amount of uncompressed original image data is enormous, it is often stored or supplied in a compressed state. Therefore, it is desirable that the data of such a reduced image is also provided in a compression encoded state, and normally,
It is re-encoded after subsampling.

In recent years, discrete cosine transform (DCT) has attracted attention as an image coding system, and the JPEG system, which has become an international standard, also uses DCT. The compression / expansion means has already spread widely all over the world by the JPEG system, and a lot of software compressed by the JPEG is about to be distributed.

For example, let us consider a case in which JPEG-encoded image data of an image that has been reduced in size by ½ is obtained from JPEG-encoded image data. In this case, in the conventional example, the JPEG-coded image data is JPEG-decoded and returned to the spatial area, and the low-pass filtering is performed to subsample the horizontal and vertical halves to obtain the spatial data of the reduced image. Thereafter, the spatial data of this reduced image is JPEG encoded. Thereby, encoded data of the reduced image can be obtained.

[0005]

In the JPEG system, a two-dimensional DCT transform of 8 pixels × 8 pixels is usually used. This DCT conversion is basically achieved by multiplying an 8 × 8 matrix twice. Therefore, in the conventional example, assuming that there are K blocks in the original image, matrix operation is performed 2K times during decoding, and K / 4 blocks are targeted during re-encoding. Will perform matrix operations twice,
A total of 5K / 2 matrix product operations are required.

In addition, from the spatial data of the original image,
A separate calculation is required to obtain a reduced image of two sizes. If filter processing is performed as pre-processing for reduction in order to improve the quality of the reduced image, the amount of calculation will be significantly increased.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above problems and to provide an encoding device and method capable of obtaining encoded data of a desired reduced image from encoded image data with less calculation. To do.

[0008]

An encoding device according to the present invention is a device for converting image code data that has been orthogonally transformed and coded in block units, and decodes the image code data into orthogonal transform coefficient data. It is characterized in that it comprises a decoding means, a synthesizing means for calculating one transform coefficient block from two or more n transform coefficient blocks, and an encoding means for re-encoding the transform coefficient block by the synthesizing means.

An encoding method according to the present invention is a method for transforming image code data that has been orthogonally transformed and coded in block units, and includes a decoding step of decoding the image code data into orthogonal transform coefficient data, and two or more. Of n transform coefficient blocks, a combining step of calculating one transform coefficient block, and an encoding step of re-encoding the transform coefficient block by the combining step.

[0010]

By the above-mentioned synthesizing means or steps, the orthogonal transform coefficient data of the image can be transformed into the orthogonal transform coefficient data of the reduced image of 1 / n without performing the inverse orthogonal transform and the subsampling. The format of the encoded image data can be converted by similar processing.

[0011]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a schematic block diagram of an embodiment of the present invention. First, the operation will be described by taking the case of reducing a JPEG encoded image as an example. Note that the JPEG encoding circuit basically outputs a DCT conversion circuit that performs DCT conversion of original image data in block units of a predetermined size, a quantization circuit that quantizes conversion coefficients by the DCT conversion circuit, and an output of the quantization circuit. It is composed of a variable length coding circuit for performing variable length coding (VLC).

Reference numeral 10 is an input terminal for JPEG encoded image data, 12 is a variable length decoding circuit for variable length decoding the encoded data from the input terminal 10, and 14 is a variable length decoding circuit 1.
2 is an inverse quantization circuit that inversely quantizes the output of 2. The output of the inverse quantization circuit 14 is DCT transform coefficient data.

16 is at least image width × 2 blocks
A memory having a memory capacity capable of storing the conversion coefficient data of the line, 18 is a memory control circuit for controlling writing and reading of the memory 16, and the memory 16 includes the inverse quantization circuit 14 under the control of the memory control circuit 18. Output (conversion coefficient data) is temporarily stored.

Reference numeral 20 denotes the memory 1 by the memory control circuit 18.
6 is a conversion coefficient synthesizing circuit for synthesizing the conversion coefficient data read from No. 6; The conversion coefficient synthesizing circuit 20 plays an important role in this embodiment, and the details of its function will be described later.

Reference numeral 22 is a quantizing circuit for quantizing the output of the transform coefficient synthesizing circuit 20, 24 is a variable length coding circuit for variable length coding the output of the quantizing circuit 24, and 26 is an output of the variable length coding circuit 24. Is an output terminal for outputting to the outside.

In the following description, for ease of understanding, the original image is composed of luminance data. The same applies to L ^* and RGB planes. Further, it is assumed that the original image is a block composed of 8 pixels × 8 pixels, and there are M blocks in the x direction and N blocks in the y direction. As a result, the image size becomes M × 8 pixels in the x direction and N × 8 pixels in the y direction, and the total number of blocks becomes M × N.

The code data input to the input terminal 10 is the JPEG code of the original image data, and is the quantized DCT.
It is a variable length code of the transform coefficient. Variable length decoding circuit 12
Is a variable length decoding of the coded data from the input terminal 10,
The quantized DCT transform coefficient data is used for the inverse quantization circuit 14
Apply to. The inverse quantization circuit 14 is a variable length decoding circuit 12
Dequantize the output of. The output of the inverse quantization circuit 4 is DCT
The conversion coefficient data is written in the memory 16 by the memory control circuit 18.

In this way, the two blocks shown in FIG.
When the DCT transform coefficient data for the line, that is, for the 2M block is written in the memory 16, the
The data of the 4 (= 2 × 2) blocks (the blocks A, B, C, and D shown in FIG. 2 at first) are read out from the left, and the transform coefficient synthesizing circuit 20 synthesizes them into one block. . Details of this combination will be described later. The quantizing circuit 22 quantizes the DCT transform coefficient data output from the transform coefficient synthesizing circuit 20, and the variable length coding circuit 24 variable length codes the output of the quantizing circuit 22. The output of the variable-length coding circuit 24 is the JPEG code of the image that is vertically and horizontally reduced to 1/2, and is output from the output terminal 26 to the outside.

After combining (and re-encoding) the transform coefficients for all of the 2M blocks stored in the memory 16, the variable-length decoded and dequantized DCT transform of the next block line is performed. The coefficient data is written in the memory 16.

Here, the processing of the circuits 12 and 14 and the processing of the circuits 20 to 24 are alternately executed so that the operation of each unit can be easily understood. However, for example, the capacity of the memory 16 is doubled. Therefore, the processing of the circuits 12 and 14 and the circuit 2
It is obvious that the processing of 0 to 24 can be executed simultaneously.

Next, the operation of the transform coefficient synthesizing circuit 20 will be described in detail by taking a one-dimensional data string as an example. Adjacent 8
The pixel arrangement is A, and the arrangement of eight pixels adjacent to the right of A is B, and the conversion coefficients a and b obtained by performing 8th-order DCT conversion on A and B are respectively a = (a ₀ , a ₁ , a ₂ , ..., A ₇ ) b = (b ₀ , b ₁ , b ₂ , ..., B ₇ ). When the basis of DCT is φ ₀ to φ ₇ , A, B,
a _i , b _i , and φ _i have the following relationship. That is,

[0023]

[Equation 1]

Here, the 16th-order waveform in which A and B are arranged is C
Then, the 16th-order wave groups δ _i and λ _i are newly defined by the following equation. That is,

[0025]

[Equation 2]

By this,

[0027]

[Expression 3] C = Σa _i σ _i + Σb _i λ _i

Let the 16th-order DCT basis be ψ ₀ to ψ _15, and C
Let c = (c ₀ , c ₁ , c ₂ , ..., C ₁₅ ) be the transform coefficient c when the 16th-order DCT transform is performed. Then, C = Σc _i ψ _i . Each element c _j is given by the following equation.

[0029]

## EQU4 ## c _j = (Σa _i σ _i + Σb _i λ _i ) ψ _j

[0030]

When a _i and b _i are put out from the equation (4),

[0031]

(Equation 5)

Of c _i thus obtained, c _{0 to} c ₇
Only C is extracted as a low frequency component of C, and this is used as a coefficient e when the vertical and horizontal 1/2 reduced region of C is converted by the 8th order DCT. Note that e = (e ₀ , e ₁ , e ₂ , ..., E ₇ ).

Using the constant d,

[0034]

(Equation 6)

It is assumed that

Among the matrix elements of the matrices P and Q of [Equation 6], if the sign of ψ is an even number, it is 8 when the coefficient of σ and λ is not 0.
The functional form is the same as the basis of the next DCT. Therefore,

[0036]

(Equation 7)

[0037] When the number of ψ is odd,
From the functional form

[0038]

(Equation 8)

It becomes

Therefore, the solution to be obtained e = (e ₀ , e ₁ , e ₂ ,
..., e ₇ ) is

[0041]

[Equation 9]

It becomes Here, M is a matrix of 4 rows and 8 columns that is uniquely determined, and d is a constant that is uniquely determined.

In this way, a = (a ₀ , a ₁ , a which has already been transformed into a coefficient by the 8th order DCT is used.

₂ , ..., A ₇ ) and b = (b ₀ , b ₁ , b
₂ , ..., b ₇ )

By applying the following equation, e = (e ₀ , e ₁ ,
e ₂ , ..., e ₇ )

Cut

[Mathematical formula-see original document] Since only a small number of operations are required, it is possible to obtain a desired result at high speed. e = (e ₀ , e ₁ , e ₂ , ...,
It should be noted that e ₇ ) is the DCT transform coefficient representing the low frequency component of the region C in which the spatial data A and B which are the bases of a and b are combined.

[0046]

The internal processing of the transform coefficient synthesizing circuit 20 will be specifically described with reference to the following equation. As shown in FIG. 3, adjacent transform coefficient blocks P, Q, R, and S are taken out. P, Q, R,
Each S is a two-dimensional DCT transformed by the 8th DCT basis. Each block P, Q, R, S is vertically cut into strips, and these are cut as shown in FIG. 4 with p ₀ to p ₇ , q ₀ to
Let q ₇ , r _{0 to} r ₇ , and s _{0 to} s ₇ .

For p _i and r _i , and q _i and s _i

[Equation 9] is actuated, and its outputs are set as t _i and u _i . i =
Perform these processes for 0-7, {t _i },
The blocks of {u _i } are named T and U as shown in FIG. The obtained T and U are laterally cut into strips, and these are set to τ ₀ to τ ₇ and ν _{0 to} ν ₇ , as shown in FIG.

For τ _i and ν _i

[Equation 9] is operated and the output is set to e _i . Obtained e ₀ ~
The 8 × 8 transform coefficient constituted by e ₇ is the solution to be obtained. It is also a block of low frequency parts within a 16x16 block that describes the same space as P, Q, R, S. That is, PQRS is reduced well (1 in each of the vertical and horizontal directions)
/ 2) is equivalent to the transform coefficient of the two-dimensional DCT based on the 8th-order DCT basis for the 8 × 8 block.

Thus, in the embodiment shown in FIG.
Without performing DCT, two-dimensional inverse transformation and image reduction processing,
With simple calculation, it is possible to directly obtain the DCT transform coefficient of the reduced image from the existing DCT transform coefficient.

In the embodiment shown in FIG. 1, the transform coefficient synthesizing circuit 2 is used.
By changing the operation of 0, various format conversions can be realized. A color image defined by the international standard system uses Y-Cb-Cr as a color space. Depending on the sampling ratio, a format such as 4: 4: 2, 4: 2: 2 and 4: 2: 0 or There are variations. In the JPEG system, 4: 2: 0 is often used. However, the format is not limited to 4: 2: 0, and other formats can be used.

Conversion between formats is a very important technique because of reduction of the amount of data. By operating the conversion coefficient synthesizing circuit 20 as follows, the chroma data of the image of the 4: 2: 2 format is applied to the embodiment shown in FIG. 1 to convert it to the 4: 2: 0 format. be able to.

An image in the 4: 2: 2 format is shown in FIG.
As shown in, the Cb and Cr data are sampled at 1/2 in the horizontal direction as compared with the Y (luminance) data. Y,
The data on the three planes of Cb and Cr are actually cut out in units of 8 × 8 blocks and DCT-encoded. Actually, there are four Y blocks, two Cr blocks, and two Cb blocks, for a total of 6 block units.

On the other hand, the 4: 2: 0 format is shown in FIG.
As shown in (4), only the chroma surface of the image of the 4: 2: 2 format is vertically sampled to 1/2.

Since the Y data of the 4: 2: 2 coded data can be used as it is, only the chroma coded data block needs to be processed. As shown in FIG. 9, the conversion coefficients of two vertically adjacent blocks P and Q are taken out, and these are vertically cut into strips as shown in FIG. 10, and each of them is converted into p ₀ to p ₇ , q _0. ~ Q ₇ .

With p _{0 to} p ₇ and q _{0 to} q ₇ as inputs

The output e _i can be obtained by operating the following equation. The arrangement of the obtained e _i as shown in FIG. 11 corresponds to the DCT transform coefficient of the original image of the blocks P and Q, which is reduced in size by half in the vertical direction. By subjecting this to variable length coding again, coded data of an image of 4: 2: 0 can be obtained.

In the above embodiment, one transform coefficient block is generated from two one-dimensional transform coefficient blocks, which corresponds to a reduction of 1/2. However, this is expanded to an arbitrary n It is also possible to generate one transform coefficient block from the transform coefficient blocks of. That is, it can be reduced to 1 / n.

Considering continuous n blocks with one block of eight adjacent data in one dimension. These are defined as Bi (original picture), and the result of Bi-order DCT conversion of Bi is bi = (bi (0), bi (1), ...
(7)). In addition, the basis of the 8th DCT transform is φ ₀ ~
φ ₇ Furthermore, the basis of the 24th-order DCT transform is ψ ₀ ~
ψ ₂₃ , and an array of eight zeros is Z = (0,0,0,0,
0,0,0,0), the 24th-order vector group σ _kj is defined by the following equation. That is,

[0058]

[Equation 10]

The entire Bi is reduced to 1/2 and the 8th-order DCT is performed.
Letting the converted conversion coefficient be e = (e ₀ , e ₁ , e ₂ , ..., E ₇ ), e can be obtained by the following procedure. That is,

[0060]

[Equation 11]

In this way, the DCT transform coefficient of the 1 / n reduced image can be obtained from the array of arbitrary n transform coefficient blocks.

In the horizontal direction with respect to the two-dimensional array of transform coefficient blocks by the two-dimensional DCT transform.

M blocks, and n blocks in the vertical direction,
Vertically and horizontally separately

By carrying out the conversion by the equation (11), it is possible to carry out scaling by an arbitrary integer fraction in the vertical and horizontal directions. For example, 4:
It is also possible to reduce the image size vertically and horizontally to 1/3 while converting from the 4: 4 format to the 4: 2: 0 format.

It goes without saying that the sampling format conversion and / or arbitrary scaling can be performed by the same processing unless the DCT conversion coefficient after conversion is limited to the 8th order.

Although the DCT has been described as an example of the orthogonal transform coding, the present invention is not limited to this, and it is also apparent that other orthogonal transform coding may be used.

[0066]

As can be easily understood from the above description, according to the present invention, it is possible to obtain the orthogonal transform coefficient of the reduced image without performing the inverse orthogonal transform and the resolution conversion in the real space. Become. As a result, the coded image data can be converted into the coded image data of the reduced image and the format of the coded image data can be converted by a simpler arithmetic process. Of course, it is possible to convert the format of the encoded image data into the encoded image data of the reduced image while converting the format of the encoded image data.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of an embodiment of the present invention.

FIG. 2 is a transform coefficient block stored in a memory 16.

FIG. 3 is an operation explanatory diagram of a conversion coefficient synthesizing circuit 20.

4 is an explanatory diagram of vectors p, q, r, s formed from blocks P, Q, R, S shown in FIG.

FIG. 5

This is the result of applying

FIG. 6 is data obtained by cutting the data shown in FIG. 5 into strips in the horizontal direction.

FIG. 7 shows image data in a 4: 2: 2 format.

FIG. 8 is image data in a 4: 2: 0 format.

FIG. 9 is a processing unit of chroma data when converting 4: 2: 2 to 4: 2: 0.

10 is data obtained by vertically cutting the conversion coefficients of blocks P and Q shown in FIG. 9 into strips.

FIG. 11 is a format conversion result.

[Explanation of symbols]

 10: Encoded image data input terminal 12: Variable length decoding circuit 14: Inverse quantization circuit 16: Memory 18: Memory control circuit 20: Transform coefficient synthesis circuit 22: Quantization circuit 24: Variable length coding circuit 26: Output Terminal

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical indication H04N 1/41 Z

Claims (8)

[Claims] 1. A device for converting image code data orthogonally coded in block units, the decoding device decoding the image code data into orthogonal transform coefficient data, and n or more transform coefficient blocks. An encoding apparatus comprising: a synthesizing unit that calculates one transform coefficient block from the above, and an encoding unit that re-encodes the transform coefficient block by the synthesizing unit. 2. The coding apparatus according to claim 1, wherein the coding means is a variable length coding means. 3. The coding apparatus according to claim 1, wherein the coding means comprises a quantizing means and a variable length coding means. 4. The coding apparatus according to claim 1, wherein the orthogonal transform coding is a discrete cosine transform. 5. A method for transforming image code data that has been orthogonally transformed and coded in block units, the decoding step of decoding the image code data into orthogonal transform coefficient data,
An encoding method comprising: a combining step of calculating one transform coefficient block from two or more n transform coefficient blocks; and an encoding step of re-encoding the transform coefficient block by the combining step. 6. The encoding method according to claim 5, wherein the encoding step is a variable length encoding step. 7. The encoding step comprises a quantization step of quantizing the data of the transform coefficient block by the synthesizing means, and a variable length encoding step of variable length encoding the result of the quantization step. The encoding method according to Item 5. 8. The coding method according to claim 5, wherein the orthogonal transform coding is a discrete cosine transform.