JP3227170B2 - Image coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding apparatus for quantizing image data and performing coding processing.

[0002]

2. Description of the Related Art In recent years, an adaptive DCT (Discrete Cosine Transform) coding method has been drawing attention as a coding method for a color image signal, and a group established as an international standardization organization for this kind of coding method. A JPEG (Joint
Photographic Expert Group
Also in the encoding method in p), the DCT encoding method is adopted.

[0003] An outline of a basic system of this type of coding system will be briefly described below.

FIG. 3 is a block diagram for explaining a schematic configuration example of a conventional encoding system using DCT transform.
(A)-(D) is a figure for demonstrating the process of the encoding system shown in FIG. Reference numeral 15 denotes an input terminal of a digital image signal to be encoded, and receives a digital image signal by raster scanning.

The image signal inputted to the terminal 15 is 8 × 8
The signal is input to the blocking circuit 16, where it is two-dimensionally (8
× 8) The image data is divided into pixel blocks each composed of pixels, and transmitted to the subsequent stage in units of the pixel blocks.

Reference numeral 17 denotes a DCT transform of the image signal from the blocking circuit 16 to obtain an (8 × 8)
Is a DCT conversion circuit that outputs the data matrix of.
That is, the image data D ₁₁ pixel block consisting to D ₈₈ as shown in FIG. 4 (A), is converted into the data matrix of X ₁₁ to X ₈₈ such shown in FIG. 4 (B) by the circuit 17.

[0007] Here, X ₁₁ is a direct current (DC) component in the horizontal direction and the vertical direction of the pixel block, that is, the average value of the pixel block. When the X ₁₁ to X ₈₈ generally as X _ij, component i has a higher frequency in the vertical direction larger have shown a component having a higher frequency in the horizontal direction as j increases.

[0008] The data matrix output from the DCT conversion circuit 17 is input to a linear quantization circuit 18. on the other hand,
The quantization matrix generation circuit 19 generates each of the DCT coefficients X ₁₁ to X ₁₁ .
Quantization matrix _W 11 to _W-88 showing the weighting of the quantization step size against us to X ₈₈ (shown in FIG. 4 (C))
, And the generation circuit 20 generates a coefficient (quantized coefficient) C. The quantization matrix W ₁₁ to _W-88 and the coefficient C is inputted to the multiplier 21. The multiplier 21 calculates (W _ij × C), and the quantization step of the linear quantization circuit 18 is determined according to the outputs Q _{11 to} Q ₈₈ of the multiplier 21.

Here, the coefficient C is a positive value.
Controls the image quality and the amount of generated data.

In practice, the linear quantization circuit 18 uses X _ij /
Q _ij is calculated and output. The output of the linear quantization circuit and G ₁₁ ~G _88. This quantized conversion data G
₁₁ ~G ₈₈ is sent in order from the DC component in the zigzag scanning circuit 22. That is, the zigzag scanning circuit 22 outputs G ₁₁
~G ₈₈ is _{G 11, G 12, G 21} , G 31, G 22, G 13, G 14,
_{G 23, G 32, G 41} ... G 85, G 86, G 77, G 68, G 78, G
₈₇ and G ₈₈ are supplied to the variable length coding circuit (VLC) 23 in this order.

In the VLC 23, for example, a DC component G
_{For 11} , a prediction value is calculated between neighboring pixel blocks, and a prediction error from the prediction value is Huffman-coded.

[0012] In addition, other than the DC component G ₁₁ AC component G ₁₂ ~
For G _88, as described above the quantization output, coded while zigzag scanning from a low frequency component to high frequency component, significant coefficients quantized output is not zero is classified into groups by the value, the group identification number And the run length of the number of invalid coefficients whose quantization output sandwiched between the immediately preceding significant coefficient is 0, and Huffman coding is performed. Then, which value in the group is equal length is determined. Add a sign.

In general, since the high frequency component in the oblique direction of the image has a low probability of occurrence, it is expected that the latter half of G _ij after the zigzag scanning is often all zero. Therefore, a very high compression rate can be expected for the variable-length code obtained in this way. If an average compression rate of about a fraction is assumed,
An image with almost no image quality degradation can be restored.

On the other hand, a transmission path generally has a predetermined transmission capacity per unit time, and when one screen must be transmitted every predetermined period, such as when transmitting a moving image, a code to be output is used. Is desired to be a fixed number of bits for each screen or pixel block.

Here, if the aforementioned coefficient C is increased, C _ij
Is increased, and the total number of bits NB of the encoded data is reduced. The relationship between the coefficient C and the total number of bits NB differs depending on the image, but is a simple decreasing function in any case, and it is known that an average image has a logarithmic curve as shown in FIG.

Therefore, a method for predicting a coefficient C0 for obtaining a desired total number of bits NB0 has been proposed by JPEG or the like. That is, encoding is first performed for a certain coefficient CI, and the total bit number nb1 of the code thus obtained is obtained. A predicted value C2 of C0 based on nb1 and C1
This can be predicted from the fact that the logarithmic curve shown in FIG. 4D passes over (C1, nb1).

By repeating this operation several times, the error code amount can be reduced to about several percent of the desired total number of bits NB0.

[0018]

However, the process of determining the value of the coefficient C0 by performing repetitive coding in this manner is a very time-consuming process, and one screen is displayed in a predetermined period like a moving image. It is not suitable for coding devices that must be transmitted. In particular, when processing an image signal having a very high bit rate such as a high-definition television signal, such processing is impossible.

Under such circumstances, the present invention provides an image coding apparatus capable of quickly and accurately obtaining a quantization step in which a code amount at the time of coding image data becomes a desired code amount. With the goal.

[0020]

An image encoding apparatus according to the present invention receives continuous image data, and controls the code amount of the image data by quantizing the image data. Generating means for generating a quantization step in which a code amount when a predetermined unit of the input image data is coded is predicted to be a desired code amount; and Encoding means for quantizing and encoding the input image data by a quantization step generated by the generating means, wherein the generating means comprises N (N is a constant of 2 or more natural numbers) codes An amount calculating means, and a predicting means for generating N quantization steps, wherein the N code amount calculating means converts the inputted image data into M-th (M is a natural number satisfying 1 ≦ M ≦ N) Set) quantum M-th code amount calculation means for obtaining an M-th code amount when quantized and coded in the step, wherein the prediction means calculates the M-th code amount from the M-th quantization step and the M-th code amount. Generating an (M + 1) -th quantization step predicted to have a code amount of, and using a quantization step predicted for image data immediately before the image data to be processed as a first quantization step, The quantization step supplied to the quantization means is an (N + 1) th quantization step generated by the prediction means.

[0021]

[0022]

[0023]

Embodiments of the present invention will be described below.

FIG. 1 is a block diagram showing the configuration of an encoding apparatus in which the present invention is applied to a transmission apparatus for transmitting a television signal as one embodiment of the present invention.

In the figure, reference numeral 1 denotes an input terminal of an analog television signal. A television signal inputted from the terminal 1 is digitized into 8 bits by an A / D converter 2, and is converted into a block circuit shown in FIG. Performs the same operation as (16 × 8)
The image data is divided into (8 × 8) pixel blocks by the blocking circuit 3 and supplied to the DCT conversion circuit 4 for each block.

The pixel data D _{11 to} D ₈₈ of each block are DC
In the T conversion circuit 4, the data is converted into data matrices X _{11 to} X ₈₈ for the frequency domain as in the case of FIG. 3 and supplied to the zigzag scanning circuit 5. The zigzag scanning circuit 5 includes:
The same operation as 22 in FIG. 3 is performed, and the DCT-transformed data matrices X _{11 to} X ₈₈ are converted into X ₁₁ , X ₁₂ , X ₂₁ , X ₃₁ , X _{31
22, X 13, X 14,} X 23, X 32, X 41 ...... X 85, X 86, X
_77, X _68, X _78, and outputs in the order of X _87, X _88.

The quantization matrix generation circuit 6 generates the above-described quantization matrices W _{11 to} W ₈₈ . However, since the data already zigzag scanning in each quantization circuit 7a~7e in this embodiment is input quantization matrix W ₁₁ to W-
_{88 is} also generated in the order corresponding to the zigzag scanning and supplied to the multiplication circuits 8a to 8e.

The prediction initial coefficient is supplied to the multiplication circuit 8a from the coefficient calculation circuit as the above-described quantization coefficient (control coefficient) C1. The prediction initial coefficient will be described later.

The multiplication circuit 8a is a quantization matrix generation circuit
6 from the quantization matrix W₁₁~ W ₈₈With the control coefficient C1
The multiplication is performed, and the result is output to the quantization circuit 7a.

[0030] Thus, quantization code G1 ₁₁ ~G1 ₈₈ by the control coefficient C1 is in the quantization circuit 7a is obtained. The quantized conversion codes G1 _{11 to} G1 ₈₈ are V
Input to LC9a.

In the present embodiment, the VLCs 9a to 9d do not output actual encoded data, and the total bit number information nb1 for each screen when the same processing as that of the VLC 23 in FIG. 3 is performed.
To nb4 only. The total bit number information nb1 output from the VLC 9a is input to the coefficient operation circuit 10.

The coefficient operation circuit 10 has VLCs 9a to 9d
, A control coefficient C0 corresponding to a desired total number of bits NB0 is predicted using the total bit number information nb1 to nb4, the initial coefficients and the outputs C1 to C4 of the arithmetic circuit 10, and C2 to C5 are output as control coefficients, respectively. I do. Here, the coefficient operation circuit 10 converts the control coefficients C2 to C5 obtained by the conversion data for one screen input to the quantization circuits 7a to 7d into the conversion data for the next one screen to the quantization circuits 7a to 7d. Output at the input timing.

On the other hand, reference numerals 11a to 11d denote circuits (1F) for delaying the output of the zigzag scanning circuit 5 by one screen (frame) period.
DL). Therefore, the control coefficient C2 output from the coefficient calculation circuit 10 is input to the multiplication circuit 8b at the timing when the conversion data for one screen used for obtaining the control coefficient C2 is input to the quantization circuit 7b. Is done.

[0034] That is, the quantization for the second time for the same screen in a quantum circuit 7b is performed, quantization code G2 ₁₁ ~G2 ₈₈ is obtained by the control coefficient C2. Conversion code G2 ₁₁ ~G2 ₈₈ wherein the quantized is input to VLC9b.

Quantization circuits 7c-7e, multiplication circuits 8c-8
e, the operations of the VLCs 9b to 9d and the 1FDLs 11b to 11d are respectively performed by the quantization circuit 7b, the multiplication circuit 8b, and the VLC 9a.
And the operation of the 1FDL 11a described above, and the predicted values of the desired control coefficients for one screen are sequentially updated by these circuits.

When the processing is performed as described above, the coefficient operation circuit 1
The predicted value C5 of the control coefficient obtained from 0 converges to a value very close to the control coefficient C0 corresponding to the desired total number of bits NB0. In this embodiment, the control coefficient C5 is supplied to the multiplying N circuit 8e as the final control coefficient C. Multiplication circuit 8e
Is supplied to a quantization circuit 7e, and the quantization circuit 7e
Then, the output of the 1FDL 11d, that is, the converted data delayed for a convenient four frame period is quantized and supplied to the VLC 9e.

The VLC 9e actually performs encoding as described with reference to FIG. 3, and outputs encoded data (DATA). The encoded data is output at a predetermined bit rate in the entry buffer 12, multiplexed with the final control coefficient C in the multiplexer 13, and output from the terminal 14 to the transmission line.

Here, the prediction initial coefficient will be described in detail.

In general, a moving image has a high correlation between screens. Therefore, adaptive DCT coding is performed to perform one screen (one frame).
When the control coefficient (quantization coefficient) is set so as to keep the total data amount of the image constant, the control coefficient has substantially the same value between the screens.

Therefore, in this embodiment, by utilizing the fact that the correlation between the screens is high, the value obtained in the previous screen is used as the control coefficient input to the multiplication circuit 8a.
The convergence of the control coefficients C2 to C5 until the optimum control coefficient is obtained can be accelerated, and a more accurate control coefficient can be obtained.

This will be described more specifically with reference to FIG.

FIG. 2A shows the values of C1 to C5 when a control coefficient for a certain screen is obtained. As shown in this figure, the control coefficients converge while oscillating to the target optimum control coefficient value.

FIG. 2B shows CI to C5 when the control coefficient of the next screen is obtained. Here, C1 input first is a control coefficient C2 of the previous screen, that is, a value of C2 in FIG. 2A (this value is called a prediction initial coefficient).
Is entered. Since the optimum control coefficient of this screen can be said to be almost the same as the previous screen, convergence is accelerated by the fact that C1 approaches the optimum control coefficient, and a value closer to the optimum control coefficient is obtained.

If the correlation between the screens is further high, this operation is repeated to obtain a higher quality image.

In this embodiment, the control coefficient is determined for each screen in order to realize extremely high-speed processing. However, if there is some margin in the processing time, the value of this control coefficient is determined. It is also possible to reduce (short) the unit period for determining.

[0046]

As described above, according to the present invention,
It is possible to quickly and accurately obtain a quantization step for setting a code amount when a predetermined unit of image data is encoded to a desired code amount.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an image encoding device as one embodiment of the present invention.

FIG. 2 is a diagram for explaining how control coefficients converge.

FIG. 3 is a block diagram for describing a schematic configuration example of a conventional encoding method.

FIG. 4 is a diagram for explaining processing of the encoding method shown in FIG. 3;

[Explanation of symbols]

 Reference Signs List 3 Blocking circuit 4 DCT transformation circuit 6 Quantization matrix generation circuit 7a-7e Quantization circuit 8a-8e Multiplication circuit 9a-9e Variable length coding circuit 10 Coefficient operation circuit 11a-11e Delay circuit 12 Entry buffer 13 Multiplexer

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-3-3479 (JP, A) JP-A-3-53666 (JP, A) JP-A-3-166887 (JP, A) Keiji Nemoto (1 other) Name), "Code amount control method of DCT coding method", Proc. Of the IEICE Autumn National Convention, The Institute of Electronics, Information and Communication Engineers, August 15, 1989, D-45, Volume 6, p. 6-45 (58) Field surveyed (Int. Cl. ⁷ , DB name) H04N ^7/ 24-7/68 H04N 1/41-1/419 H03M 7/30

Claims (1)

(57) [Claims] 1. An image coding apparatus for inputting continuous image data, controlling the code amount of the image data by quantizing the image data, and coding the image data, comprising: Generating means for generating a quantization step in which a code amount when a predetermined unit of encoded image data is encoded is predicted to be a desired code amount, and the quantization step generated by the generation means Encoding means for quantizing and encoding the image data, wherein the generating means generates N (N is a natural number of 2 or more) code amount calculating means and N quantization steps A prediction unit, wherein the N code amount calculation units quantize and encode the input image data in an Mth (M is a set of natural numbers satisfying 1 ≦ M ≦ N) quantization step; M-th sign The prediction means comprises: an (M + 1) th quantization step for predicting the desired code amount from the Mth quantization step and the Mth code amount. The quantization step generated and used as a first quantization step is a quantization step predicted for the image data immediately before the image data to be processed, and supplied to the encoding means, An image coding apparatus, wherein the generated (N + 1) th quantization step is performed.