JPH1023413A - Encoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an encoding device of a small device scale by executing pipeline processing in each divided picture. SOLUTION: A line sequential output means 1 outputs picture data in the order of scanning lines. A 1st switching means 2 divides the picture data in each fixed scanning line order to switch an output destination. Each of storage means 3a to 3d stores the divided picture data in each scanning line order. Each of pixel block constituting means 5a to 5c reads out pixels from respective storage means 3a to 3d in fixed order to constitute a pixel block. Each of 2nd switching means 4a to 4c switches connections between the means 3a to 3d and the means 5a to 5c in fixed order. An encoding amount calculating means 6 calculates an encoding amount from a quantization coefficient. An optimum quantization control coefficient calculating means 7 calculates an optimum quantization control coefficient based on the encoding amount. An encoding means 8 encodes a pixel block by using the optimum quantization control coefficient.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an encoding device,
In particular, the present invention relates to an encoding device that encodes image data.

[0002]

2. Description of the Related Art An orthogonal transform coding method is known as a method for efficiently coding an image having gradations. In particular, JPEG (Jo) using DCT (Discrete Cosine Transform)
intPhotographic Expert Gr
up) or MPEG (Moving Picture)
A method called “Expert Group” is an international standard method.

[0003] In DCT, since the power of image data is mostly concentrated on low frequency components in an area that does not include edges, a high compression rate can be obtained without affecting the image quality even if the high frequency components are coarsely quantized. Can be On the other hand, when an edge or the like is included in the pixel block, the power is dispersed even in high frequency components, so that the compression ratio decreases.

When changing the compression ratio or image quality,
A process of multiplying the set quantization matrix by a constant value to uniformly increase or decrease the quantization step width is performed. This process is called quantization scaling. Further, the constant value by which each element is multiplied is called a scaling factor or a quantization control coefficient, but is hereinafter referred to as a quantization control coefficient.

[0005] As described above, since the coding compression rate using DCT depends on the image content, it is necessary to keep the code amount of the image data constant. As a method for making the code amount of image data constant, for example, there is a method of obtaining a quantization control coefficient converging to a target code amount as disclosed in Japanese Patent Application Laid-Open No. Hei 4-233373.

FIG. 20 is a diagram showing the configuration of the above-mentioned prior art. An A / D unit 100 for analog-to-digital conversion of an input image signal into digital image data; a first encoding unit 101 for encoding image data with a quantization control coefficient q1 to calculate a code amount b1; A second encoding unit 102 that encodes the image data with a quantization control coefficient q2 to calculate a code amount b2, a delay unit 103 that delays image data for a fixed time, a target code amount b0 from the code amounts b1 and b2.
Calculation unit 1 for obtaining quantization control coefficient q0 so as to converge to
04 and the quantization control coefficient q0 determined by the arithmetic unit 104
And a third encoding unit 105 that encodes the delayed image data.

The input analog image signal is converted into digital image data by the A / D unit 100. The image data is input to the first encoding unit 101, and is encoded with the quantization control coefficient q1 to calculate a code amount b1. Similarly, the image data is input to the second encoding unit 102, and is encoded with the quantization control coefficient q2 to calculate the code amount b2.
The arithmetic unit 104 linearly interpolates the combination of the quantization control coefficient q1 and the code amount b1, and the combination of the quantization control coefficient q2 and the code amount b2,
The quantization control coefficient q0 is determined so as to converge on the target code amount b0. Then, the image data is input to the third encoding unit 105 after being delayed by the delay unit 103 for a predetermined time, and is encoded with the quantization control coefficient q0 determined by the arithmetic unit 104.

In such a conventional technique, encoding is performed by delaying the A / D converted image data by about one frame in a delay unit. In addition, at least two encoding units are parallelized, and a code amount when encoding is performed with two types of quantization control coefficients at the same time is calculated, and a quantization control coefficient serving as a target code amount is determined.

[0009]

However, in the above-described prior art, it is necessary to provide a memory in the delay unit. The memory capacity is, for example, ISO A4 size (210 [mm] × 297 [m
m]), this is set to 8 pixels per pixel.
[Bit] and a resolution of about 16 [pixels / mm], 210 × 16 × 297 × 16 = 15,966,72
This is 0-byte data. Also, the data amount when inputting in full color is three times as large as 15,966,72.
0 [pixel] × 3 [byte / pixel] = 47,900,1
It becomes a huge data of 60 bytes.

Therefore, when a plurality of encoding units encode in parallel to determine a quantization control coefficient that converges to a target code amount, a memory that only delays such a high-definition image with a large amount of data for a predetermined period. Is disadvantageous in that the size of the apparatus increases.

In addition, when no memory is provided, there is a problem in that it is necessary to repeat coding, which increases the amount of calculation and the processing time. The present invention has been made in view of such a point, and an object of the present invention is to provide an encoding device having a small device scale by performing parallel processing for each divided image.

It is another object of the present invention to provide an encoding apparatus for determining a quantization control coefficient which converges to a target code amount in one scan.

[0013]

In order to solve the above-mentioned problems, in an encoding apparatus for encoding image data, a line-sequential output means for outputting image data in a scanning line-sequential manner; First switching means for switching the output destination by dividing line by line, accumulating means for accumulating the image data divided by scanning line, and reading pixels from the accumulating means in a certain order A second block that switches the connection between the storage block and the pixel block block in a fixed order;
Switching means, switching control means for performing switching control of the first and second switching means, and a quantization coefficient which encodes the pixel block and quantizes it with a quantization characteristic determined by at least two quantization control coefficients. Calculating a code amount from the quantized coefficient, and an optimal quantization control coefficient calculating unit that calculates an optimal quantization control coefficient that converges to a target code amount based on the code amount. Encoding means for encoding the pixel block using the optimal quantization control coefficient.

Here, the line-sequential output means outputs image data in a scanning line sequence. The first switching unit switches the output destination by dividing the image data for each predetermined scanning line.
The storage unit stores the image data divided for each scanning line. The pixel block forming unit reads out the pixels from the storage unit in a predetermined order to form a pixel block. Second
Switching means switches the connection between the accumulation means and the pixel block forming means in a certain order. The switching control means performs switching control of the first and second switching means. The code amount calculating means encodes the pixel block, quantizes the pixel block with a quantization characteristic determined by a quantization control coefficient to obtain a quantized coefficient, and calculates a code amount from the quantized coefficient. The optimum quantization control coefficient calculation means calculates an optimum quantization control coefficient that converges on the target code amount based on the code amount. The encoding unit encodes the pixel block using the optimal quantization control coefficient.

[0015]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the encoding apparatus. The encoding apparatus includes a line-sequential output unit 1 that outputs image data in a scanning line sequence, a first switching unit 2 that switches an output destination of image data, and storage units 3a and 3b that store image data for a plurality of lines. 3c, 3d and second switching means 4a, 4b for switching the output from the storage means,
4c, pixel block forming means 5a, 5b, 5c forming a pixel block, a code amount calculating means 6 for calculating a code amount, an optimal quantization control coefficient calculating means 7 for calculating an optimal quantization control coefficient Q0, An encoding unit 8 that encodes image data with the optimal quantization control coefficient Q0, and a switching control unit 9 that performs switching control of the first switching unit 2 and the second switching units 4a, 4b, and 4c. You. The code amount calculating means 6 includes a first code amount calculating means 6a for outputting the code amount B1, and a second code amount calculating means 6 for outputting the code amount B2.
b.

Next, a description will be given of high-speed processing in which an input image is divided and processing of each divided image is performed in a pipeline manner. FIG. 2 is a diagram illustrating a divided image. The divided image is obtained by dividing the input image 10 into a band-like region composed of a plurality of scanning lines. Each divided image 1a to 1f
.. Are divided so as to have the same number of scanning lines.

FIG. 3 shows a case where four storage means are provided.
FIG. 11 is a diagram showing a flow of pipeline processing performed for each divided image. In the following description, the storage means will be referred to as a block line buffer. The number in parentheses in the table is the number of code amount calculations. First, in the period A, image data corresponding to the divided image 1a in FIG. 2 is read into the block line buffer 3a. Subsequently, in the period B, the first preliminary coding is performed on the divided image 1a of the block line buffer 3a, and the first code amount is calculated.
The preliminary coding will be described later. In parallel with this, the divided image 1b is read into the block line buffer 3b.

Further, in the period C, the second preliminary coding is performed on the divided image 1a of the block line buffer 3a, and the second code amount is calculated. At this time, the first preliminary coding is performed on the divided image 1b of the block line buffer 3b, and the divided image 1c is input to the block line buffer 3c.

In the period D, the block line buffer 3a
Is encoded using the quantization control coefficient estimated from the result of the second preliminary encoding described above, and encoded data is output. At this point, a series of processes on the divided image 1a stored in the block line buffer 3a is completed, and in the next period E, a new divided image is read into the block line buffer 3a.

By continuing the same operation, encoding is performed by controlling the input image to a predetermined code amount in units of divided images. Also, since a series of processes required for code amount control overlap each other, the time required for encoding can be made substantially equal to the time required for one-pass encoding process.

Next, the operation of inputting image data to the block line buffer will be described. FIG. 4 is a diagram illustrating a state in which image data is input to the block line buffer. The block line buffer 3 has a capacity equal to or larger than the maximum number of pixels that can be scanned by the line sequential output means 1 in the main scanning direction. In the sub-scanning direction, the number of lines required to configure one pixel block is provided. In the figure, there are eight lines.

FIG. 5 is a diagram showing how image data is output from the block line buffer. Image data output from the block line buffer 3 is read out in the block scan order as shown in the figure. Then, a pixel block is formed in the pixel block forming means 5a, 5b, 5c. The pixel block is composed of 8 × 8 pixels.

Next, the operation of the first and second switching means will be described. FIG. 6 is a diagram illustrating operations of the first switching unit and the second switching unit. Further, the divided image is referred to as a block line in the following description. Introduction Block Line 1
The image data a is read by the line-sequential output unit 1 and stored in the block line buffer 3a connected by the first switching unit 2.

When reading the subsequent block line 1b, the output destination of the first switching means 2 has been switched to the block line buffer 3b, so that the image data is stored in the block line buffer 3b. At this time, since the second switching means 4a is connected to the block line buffer 3a, the read operation of the image data of the first block line already stored is performed by the block line buffer 3a.
b can be executed in parallel with the write operation.

Similarly, as for the image data of the subsequent block lines, writing and reading are executed in parallel in block line units as shown in FIG. Further, the switching control means 9 controls the switching of the connection between the first switching means 2 as described above and the second switching means 4a, 4b, 4c.

As described above, the block line buffer is switched by the switching means, and the pixel block forming means performs the pipeline-like parallel processing of forming the pixel blocks.

Thus, parallel processing of image data can be executed, and even for a large-capacity image, high-speed encoding processing can be performed without requiring a memory for storing the entire original. Next, a description will be given of a code amount calculating means for performing pre-encoding. Since the first code amount calculating means and the second code amount calculating means have the same configuration, they will be described as code amount calculating means. FIG. 7 is a diagram showing the configuration of the code amount calculation means. The code amount calculation unit 601 is an orthogonal transformation unit (hereinafter, referred to as DCT) 601.
, Quantization means 603, and variable length coding means (hereinafter VL)
C. ) 600. DCT601
Divides image data into pixel blocks and performs orthogonal transformation on each pixel block to determine a transform coefficient.

FIG. 8 is a diagram schematically showing a transform coefficient matrix after DCT. As shown in the figure, it is composed of 8 × 8 elements corresponding to the horizontal and vertical frequency components. The DC coefficient 601a located at the upper left corresponds to the average luminance value of the pixel block before DCT. On the other hand, the other elements in the matrix are AC coefficients.

The quantizing means 603 quantizes the transform coefficient with a quantization matrix to obtain a quantized coefficient. The quantization matrix is set in the quantization table 602 after being scaled by the quantization control coefficients Q1 and Q2. The VLC 600 calculates the code amount B based on the quantization coefficient.
1. Calculate B2.

Next, the principle of determining the optimum quantization control coefficient for converging to the target code amount will be described in detail. It is known as an empirical rule that the logarithms of the quantization control coefficient Q and the code amount B can be approximated using a linear function represented by the following equation.

[0031]

(Equation 1)

Here, if the constant a is the same encoding,
The value of the constant b is assumed to have a constant distribution depending on the image, regardless of the image.

FIG. 9 is a diagram showing the relationship between the quantization control coefficient Q and the code amount B on the logarithmic axis. The abscissa represents logQ of the logarithm of the quantization control coefficient, and the ordinate represents logB of the logarithm of the code amount B. As can be seen from the figure, the logarithmic relationship between the quantization control coefficient Q and the code amount B is a linear function. Therefore, when encoding one image X, if it is possible to obtain each code amount B by encoding with a quantization matrix scaled by two types of quantization control coefficients Q, the constants a, b can be obtained. Then, if a and b are obtained, the optimum quantization control coefficient for converging to the specified target code amount B0 is calculated by the equation (1).
Can be obtained from

That is, two quantization control coefficients Q1, Q
If the code amounts obtained from Equation 2 are B1 and B2, Equation (1)
From

[0035]

(Equation 2)

[0036]

(Equation 3)

Is as follows. Then, by replacing B in equation (1) with the target code amount B0 and substituting a and b in equations (2) and (3) into equation (1), the optimum quantization control coefficient Q
0 can be obtained.

[0038]

(Equation 4)

As described above, in the present embodiment,
Preliminary encoding is performed at least twice on each of the divided images to determine the relationship between the quantization control coefficient and the generated code amount. Is calculated and encoding is performed.

The present embodiment basically aims at controlling each divided image so as to have the same code amount. However, the image data included in the divided image is originally generated code such as the margin of the original. In the case of an area with a small amount, the optimal quantization control coefficient may be excessively reduced. This means that, when a divided image having a large generated code amount and a divided image having a small generated code amount are adjacent to each other, the optimum quantization control coefficients calculated for both are greatly different, and the two divided images are divided into two. Visual disturbances, such as discontinuities in boundaries, may occur in decoded images.

For this reason, the range for controlling the optimal quantization control coefficient is set in advance as an upper limit and a lower limit, and
If the optimum quantization control coefficient converges outside this range, the closer one is adopted as the optimum quantization control coefficient.

Next, the VLC will be described in detail. FIG. 10 is a diagram showing the configuration of the VLC. VLC converts scan order of a matrix of quantization coefficients and outputs DC coefficients and one-dimensional AC coefficients.
0, a delay unit 61 for delaying the DC coefficient by one pixel block period, a subtractor 62 for subtracting the DC coefficient of one pixel block before and the DC coefficient of the current pixel block and outputting difference information, Is classified into a predetermined group according to its size, and a group determination unit 63a that outputs an additional bit for identifying a group number and a numerical value in the group
And code length reading means 64a for reading the length of the variable length code word corresponding to the group number from the DC coefficient code table 65 and adding the length of the additional bit to output the code length information.
And invalid coefficient determining means 66 for determining whether or not the input quantized coefficient is invalid (having a value of 0) and separating the two.
A zero run counter 67 that counts the continuous length of the invalid coefficient and outputs the length, and classifies the input valid coefficient into a predetermined group according to its size, and identifies a group number and a numerical value in the group. And an AC coefficient code table 68. The group determining means 63b outputs an additional bit for calculating the length of the variable length codeword determined from the continuous length of the invalid coefficient and the group number indicating the size of the effective coefficient following the invalid coefficient.
And a code length reading means 64b for adding the length of the additional bits to output the code length information, and an adder 69 for cumulatively adding the code lengths and calculating the code amount B per image.
And

Next, the operation of calculating the code amount B in the VLC will be described. First, the scan conversion means 60 separates the quantized coefficients into DC coefficients and AC coefficients. FIG.
FIG. 4 is a diagram illustrating a state of zigzag scanning performed by a scan conversion unit. The AC coefficient excluding the DC coefficient 601a is changed from a two-dimensional array to one by a zigzag scan as shown.
Converted to a dimensional array.

Hereinafter, the processing for the DC coefficient will be described. The DC coefficient separated by the scan conversion means 60 is stored in the delay unit 61. Then, the DC coefficient before one pixel block stored in the delay unit 61 is output.

The subtractor 62 calculates the difference between the DC coefficient one block before and the current DC coefficient, and outputs the result as difference information. The group determination unit 63a outputs a group number (SIZE) to which the input differential signal belongs, and an additional bit for identifying which numerical value in the group the differential signal corresponds to. FIG. 12 is a diagram illustrating grouping of differences between DC coefficients. For example, if the input difference signal (DC coefficient difference 63a-2) is 4, SIZE6
3a-1 is 3, and the additional bit 63a-3 is 3.

The code length reading means 64a reads the length of the variable length code word corresponding to the input SIZE from the DC coefficient code table 65. FIG. 13 shows a DC coefficient code table.
The DC coefficient code table 65 includes a SIZE 65a and a code word 65.
b and the code length 65c. For example, SIZE65
If a is 3, the code length 65c is 3.

In the case of a normal encoding process, SIZE
And a corresponding code word (bit pattern) and a code length need to be set, but in the present embodiment, a code word is not necessary and only SIZE and a code length may be set.

The length of the variable-length code word determined as described above and the length of the additional bit are added to output the code length of the DC component in one pixel block. Next, the processing for the AC coefficient will be described. Invalid coefficient determination means 66
Determines whether the value of the AC coefficient output as a one-dimensional coefficient sequence from the scan conversion unit 60 is invalid (0).

If the value is zero, the zero run counter 67 counts its continuous length (RUN). If an AC coefficient whose value is valid (not 0) is detected, the AC coefficient is output to the group determination unit 63b.

The group determining means 63b outputs a group number (SIZE) and an additional bit for identifying which numerical value in the group the detected effective coefficient corresponds to, as in the case of the DC coefficient. FIG. 14 is a diagram illustrating grouping of AC coefficients. For example, if the input AC coefficient 63b-2 is 8, the SIZE 63b-1 is 4, and the additional bit 63b-3 is 4.

When the effective coefficient is detected, the code length reading means 64b corresponds to the combination of the numerical value (RUN) counted by the zero run counter 67 and the SIZE determined by the group judging means 63b. Determine the variable length codeword.

FIG. 15 shows an AC coefficient code table. The AC coefficient code table is a combination of RUN and SIZE (RUN / SIZ
E) Consists of 68a, codeword 68b, and code length 68c. However, as in the case of the DC coefficient code table, the code word may not be set.

FIG. 16 shows a two-dimensional code table composed of RUN and SIZE. This code table 64c
Is a table for forming a combination of (RUN / SIZE) 68a in FIG.

As described above, the code length reading means 6
4b reads the length of the variable-length codeword corresponding to the combination of RUN and SIZE, adds it to the additional bit length, and outputs the codeword length of the AC coefficient.

The adder 69 comprises a code length reading means 64
The code length B of the entire image is calculated by accumulatively adding the code lengths a and 64b output over the entire image. With the above configuration and operation, the image can be encoded to the target code amount B0 specified in advance.

FIG. 17 is a diagram showing the structure of the encoding means. The encoding means 8 includes a DCT 601b and a quantization means 6
03b and the VLC 600b. The quantization means 603b quantizes the transform coefficient with a quantization matrix to obtain a quantized coefficient. After the quantization matrix is scaled by the optimal quantization control coefficient Q0, it is set in the quantization table 602b. And VLC600
b outputs coded data based on the quantized coefficients.

The coding means 8 sets a quantization matrix or an optimum quantization control coefficient to be used for coding from now on at the head of the block line before starting coding of one block line. , The quantization matrix can be changed at the time of decoding. Also,
an optimal quantization control coefficient for the (n-1) th block line;
The difference between the optimum quantization control coefficients of the n-th block line may be used.

The information instructing such a change of the quantization matrix can be set by using the same method as the marker code in JPEG. In JPEG, control information called a marker code is defined for setting the optimum quantization control coefficient and the synchronization information, and can be set in the code data.

In this method, FF (H) is inserted into the code data so as to match a byte boundary, followed by a 1-byte marker code having a value other than 0. If FF is generated as code data, 00
By adding (H), it is possible to distinguish the marker code from the code data.

In this embodiment, it is assumed that the quantization matrix is changed by using an undefined marker code of JPEG. In addition, some JPEG marker codes have variable-length fields in addition to those used alone, such as those indicating the start and end of data. It is assumed that the marker code used in the present embodiment has a variable-length field for setting the content of change of the quantization matrix.

Further, in order to set the information instructing the change of the quantization matrix so as to match the byte boundary,
The encoding means 8 adds a fill bit after outputting the encoded data so that each block line ends at a byte boundary.
This not only facilitates the detection of information instructing the change of the quantization matrix, but also allows extraction of code data corresponding to a specific block line from the code data.

Next, processing for preventing the pipeline operation from being broken will be described. Generally, the input operation of a document by a scanner or the like is performed at a constant speed, so that the time required for the input processing of the divided image can be considered to be constant. Also,
The processing time of the final encoding process can be made substantially constant by controlling the code amount. Therefore, control can be performed within the time required for image input processing by setting a compression ratio as a control target.

On the other hand, in the pre-encoding process, since the encoding process is performed while the original characteristics of the divided image are unknown, all the processes may not be completed within a predetermined time. In such a case, the preliminary coding is not performed on all the image data in the block line buffer, and a partial area is to be subjected to the preliminary coding. As a partial area,
Odd-numbered or even-numbered pixel blocks may be used alone, or in the sense that a document margin portion with a small amount of information is excluded,
The pixel block at the center of the block line buffer may be used.

Also, by notifying the end of processing by the line sequential output means to the code amount calculating means, the pre-encoding processing is forcibly ended, and the number of blocks for which the pre-encoding processing has been completed at that time The code amount of the entire block line may be estimated from the code amount and the code amount.

Next, a second embodiment of the present invention will be described. FIG. 18 is a configuration block diagram of an encoding device according to the second embodiment. The encoding apparatus includes a line sequential output unit 1 for outputting image data in a scanning line sequence, a first switching unit 2a for switching an output destination of image data, and storage units 3a, 3b for storing image data for a plurality of lines. 3c,
A second switching unit 4c for switching the output from the storage unit,
4d, pixel block forming means 5a and 5b constituting a pixel block, approximate code amount calculating means 6c for outputting code amount B1 and code amount B2, and optimal quantization control coefficient for calculating optimal quantization control coefficient Q0. Calculating means 7, coding means 8 for coding the image data with the optimal quantization control coefficient Q0,
A switching control unit 9a for performing switching control of the first switching unit 2a and the second switching units 4c and 4d.

Next, the approximate quantization coefficient will be described.
An essential system called a baseline system of JPEG which is an international standard system of still image coding is a system using orthogonal transform coding. The quantization process in JPEG is represented by the following equation.

[0067]

(Equation 5)

[0068]

(Equation 6)

Here, F 1 (i, j) is a quantization coefficient,
(I, j) represents a transform coefficient, and Q (i, j) represents a quantization matrix. i and j correspond to the positions of the elements in the individual matrices. [] Is an operator representing integer conversion.

A quantization coefficient F2 (i, j) obtained by quantizing a transform coefficient with a quantization matrix scaled by a quantization control coefficient Q = 2, and a quantization control coefficient Q = 1
Approximate quantized coefficient F2-1 (i, i) obtained by quantizing the transform coefficient by a predetermined quantization matrix scaled by
The relationship with j) can be approximately expressed by the following equation.

[0071]

(Equation 7)

Here, ≫ indicates a bit shift operation in the right direction and ≪ indicates a bit shift operation in the left direction, and the difference between both sides is due to rounding (rounding).

Therefore, the approximate quantization coefficient F2-1 (i,
j) is coded by a quantization coefficient F2 (i,
j) can be considered to be approximately equal to the code amount at the time of encoding.

As described above, the code amount obtained by performing an N-bit right shift process on the quantization coefficient obtained from the quantization matrix and the code amount obtained when the same quantization matrix is scaled by 2 ^N are obtained. In the second embodiment, the optimum quantization control coefficient is obtained by utilizing the fact that the error can be approximated within the range of the error caused by the rounding processing of the quantization.

As described above, the encoding apparatus according to the second embodiment uses the basic matrix (quantization control coefficient Q =
The code amount B1 obtained from 1) and the quantization control coefficient Q = 2
Code amount B2 obtained from the approximate quantized coefficient corresponding to
Are used to determine the constants a and b from the above equations (2) and (3). Further, the optimum quantization control coefficient Q0 is determined by substituting the constants a and b and the target code amount B0 into the equation (4). Therefore, the optimal quantization control coefficient can be determined without performing encoding in parallel by a plurality of encoding units, so that the device scale can be reduced.

In the above description, the approximate quantized coefficient is obtained by shifting the quantized coefficient to the right by one bit. However, the N-bit shift processing (where N
Is an integer, right shift is positive and left shift is negative)
It does not matter. In the case of N-bit shift processing, this corresponds to a quantization coefficient scaled by a quantization control coefficient Q = ^2N .

Next, the configuration of the approximate code amount calculating means will be described. FIG. 19 is a diagram showing the configuration of the approximate code amount calculation means. The approximate code amount calculation means includes a DCT 601, a quantization means 603, an approximate quantization coefficient calculation means 604,
VLCs 600 and 600a. The quantization means 603 quantizes the transform coefficient with a quantization matrix,
Find the quantization coefficient. After the quantization matrix is scaled by the quantization control coefficient Q1, the quantization table 6
02 is set. Also, the approximate quantization coefficient calculation means 60
Reference numeral 4 calculates an approximate quantized coefficient by performing a shift process on the quantized coefficient output from the quantizing means 603.

As described above, in the second embodiment, four block line buffers are required in comparison with the first embodiment, and three second block line buffers are required in the second embodiment. Since the number of means can be reduced to two, the device configuration can be simplified.

Further, also in the second embodiment, it is not necessary to repeat the coding in the coding unit, and the amount of calculation and the processing time can be reduced.

[0080]

As described above, the coding apparatus of the present invention switches the block line buffer in a fixed order, and performs the preliminary coding and the coding with the optimum quantization control coefficient in parallel processing. And This eliminates the need for a large-capacity memory for storing the entire document even for large-capacity images.

Further, the coding apparatus of the present invention has a configuration in which one coding unit calculates a code amount corresponding to a plurality of quantization control coefficients. Thereby, the optimal quantization control coefficient converging to the target code amount can be calculated with a small amount of calculation, and the processing time is shortened.

[Brief description of the drawings]

FIG. 1 is a configuration block diagram of an encoding device according to the present invention.

FIG. 2 is a diagram showing a divided image.

FIG. 3 is a diagram showing a flow of pipeline processing performed for each divided image when four storage units are provided.

FIG. 4 is a diagram illustrating a state in which image data is input to a block line buffer.

FIG. 5 is a diagram illustrating a state in which image data is output from a block line buffer.

FIG. 6 is a diagram illustrating operations of a first switching unit and a second switching unit.

FIG. 7 is a diagram illustrating a configuration of a code amount calculation unit.

FIG. 8 is a diagram schematically illustrating a transform coefficient matrix after DCT.

FIG. 9 is a diagram showing a relationship between a quantization control coefficient Q and a code amount B on a logarithmic axis.

FIG. 10 is a diagram showing a configuration of a VLC.

FIG. 11 is a diagram showing a state of zigzag scanning performed by a scan conversion unit.

FIG. 12 is a diagram illustrating grouping of differences between DC coefficients.

FIG. 13 shows a DC coefficient code table.

FIG. 14 is a diagram showing grouping of AC coefficients.

FIG. 15 shows an AC coefficient code table.

FIG. 16 illustrates a two-dimensional code table including RUN and SIZE.

FIG. 17 is a diagram illustrating a configuration of an encoding unit.

FIG. 18 is a block diagram illustrating a configuration of an encoding device according to a second embodiment.

FIG. 19 is a diagram showing a configuration of an approximate code amount calculation unit.

FIG. 20 is a diagram showing a configuration of a conventional technique.

[Explanation of symbols]

 1 line-sequential output means 2 first switching means 3a, 3b, 3c, 3d storage means 4a, 4b, 4c second switching means 5a, 5b, 5c pixel block construction means 6 code amount calculation means 6a first code amount Calculation means 6b Second code amount calculation means 7 Optimum quantization control coefficient calculation means 8 Encoding means 9 Switching control means

Claims (13)

[Claims] 1. An encoding apparatus for encoding image data, comprising: line-sequential output means for outputting image data in a scanning line sequence; and switching the output destination by dividing the image data for each predetermined scanning line sequence. 1 switching means, accumulation means for accumulating the image data divided for each scanning line in sequence, pixel block construction means for reading out pixels in a predetermined order from the accumulation means to constitute a pixel block, A second switching unit for switching a connection between the unit and the pixel block forming unit in a predetermined order; a switching control unit for performing switching control of the first and second switching units; Code amount calculating means for quantizing with a quantization characteristic determined by two quantization control coefficients to obtain a quantized coefficient, and calculating a code amount from the quantized coefficient; An optimal quantization control coefficient calculating unit that calculates an optimal quantization control coefficient that converges to a signal amount, and an encoding unit that encodes the pixel block using the optimal quantization control coefficient. Encoding device. 2. The encoding apparatus according to claim 1, wherein said storage means stores the image data for at least eight scanning lines. 3. The optimum quantization control coefficient calculation means sets an upper limit value and a lower limit value of the optimum quantization control coefficient, and sets the optimum quantization control coefficient out of a range between the upper limit value and the lower limit value. 2. The encoding apparatus according to claim 1, wherein, when the value converges to the value, the smaller one of the upper limit value and the lower limit value is output. 4. The encoding means, when encoding the image data, indicates a quantization characteristic at the top of code data obtained by dividing and encoding the image data for each predetermined scanning line sequence. The encoding device according to claim 1, wherein information is added. 5. The encoding apparatus according to claim 1, wherein said encoding means assigns a fill bit so that the end of the encoded data obtained by encoding the image data is a byte boundary. 6. The code amount calculation unit performs orthogonal transformation for each of the pixel blocks to obtain a transformation coefficient, and quantizes the transformation coefficient with a quantization characteristic determined by a first quantization control coefficient. First quantizing means for obtaining the quantized coefficient by a first variable length coding means, and first variable length coding means for performing a variable length coding on the quantized coefficient to calculate the first code amount. Code amount calculating means, the orthogonal transform means,
A second quantization unit that quantizes the transform coefficient with a quantization characteristic determined by a second quantization control coefficient to obtain the quantized coefficient, and performs the variable length coding on the quantized coefficient to obtain the second quantized coefficient. 2. The encoding apparatus according to claim 1, further comprising: a second variable-length encoding unit configured to calculate a code amount of the second variable length encoding unit; 7. The first and second code amount calculating means,
The encoding device according to claim 6, wherein the first and second code amounts are calculated using the same variable length code table. 8. The encoding apparatus according to claim 6, wherein said orthogonal transform means is a discrete cosine transform. 9. The optimal quantization control coefficient calculation means calculates a logarithm of the first and second quantization control coefficients and the first and second code amounts using a linear function including 7. The encoding apparatus according to claim 6, wherein the optimal quantization control coefficient is determined. 10. The code amount calculation unit includes a quantization unit that is determined by the orthogonal transformation unit, the first quantization unit, the first variable length coding unit, and the second quantization control coefficient. Characteristics, an approximate quantized coefficient calculating means for calculating an approximate quantized coefficient approximately equal to a quantized coefficient obtained by quantizing the transform coefficient, and applying the variable length coding to the approximate quantized coefficient. 2. A coding apparatus according to claim 1, further comprising a variable-length coding calculating means for calculating a second code amount by performing the calculation. 11. The approximate quantization coefficient calculation means sets the second quantization control coefficient to 2 ^N and quantizes the quantization by a quantization characteristic determined by the first quantization control coefficient. The encoding apparatus according to claim 10, wherein the approximate quantization coefficient is calculated by performing an N-bit shift process on the coefficient. 12. The encoding apparatus according to claim 10, wherein said orthogonal transform means is a discrete cosine transform. 13. The optimal quantization control coefficient calculating means,
The first quantization control coefficient and the approximate quantization coefficient,
11. The encoding apparatus according to claim 10, wherein the optimal quantization control coefficient is determined by a linear function including a parameter having a logarithm of the first and second code amounts.