JPH07212757A - Picture compression coder - Google Patents

Abstract

PURPOSE:To enhance the accuracy of code quantity control without increasing the scale of hardware by shifting a cross reference characteristic between a code quantity and a scale factor based on the number of components of picture data. CONSTITUTION:The number of components for an input picture is added to a compression rate as a parameter for range setting of a scale factor alphai of an MPU 33 to be multiplied by a quantization table 32 and the scale factor alphai is shifted depending on the number of components to decide the scale factor alphai corresponding to the code quantity and to implement quantization. Furthermore, a byte stuffing generated quantity is added to each output of code quantity calculation circuits 36-39 by adder circuits 40-43 to obtain a code quantity of each group corresponding to each scale factor and a scale factor alphat corresponding to a set code quantity Nt is calculated by a calculation circuit 46 based on the relation with each scale factor to prevent a set value overflow of code quantity by the byte stuffing processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement of an image compression coding apparatus for performing compression coding processing on a digitized video signal.

[0002]

2. Description of the Related Art As is well known, in recent years, along with the digitization of video equipment, the digitization of signal processing in cameras has been promoted. Along with this, the development of a system for recording a digitized video signal as it is on a recording medium in a digital form has been actively carried out. An example of this system is an electronic still camera. As this electronic still camera, a device that analog-records a video signal on a floppy disk is already on the market, but recently, a device that digitally records a video signal on a memory card has been announced. In this case, in order to efficiently use the limited memory capacity of the memory card, image data compression technology is essential.

As a compression means for image data that has already been announced, DPCM (DefferencialPulse Code Modulat)
Ion) method and orthogonal transform method, but at present,
It is said that data compression using DCT (Discrete Cosine Transform), which is one of orthogonal transform methods, is advantageous because there is a strong movement to increase the compression rate. Also,
ISO / C is the international standardization for compression encoding of still images.
JPEG (Joint Photograph) which is a subordinate organization of CITT
ic Expert Group), and the most used baseline method among them is also a compression coding method using DCT.

By the way, when an image is compression-encoded, the code amount generally increases with respect to a fine pattern, while the code amount decreases with a smooth image. Therefore, since the code amount at the time of compression changes for each image, there arises a problem that it becomes difficult to manage the recording capacity when recording on the recording medium. Since nothing is defined in this respect in the above-mentioned baseline method, in actual operation, by adding or subtracting the value of the quantization table used at the time of quantization, a so-called code for making the final code amount constant It has become common to control the quantity.

In this case, however, considering the hardware scale, the number of quantization tables that the system can have is limited, so a few typical ones are set for each set code amount. It is common to keep. Therefore, depending on the content of the image data, some problems remain in the accuracy of the code amount after control.

FIG. 8 shows a conventional image compression coding apparatus adopted in an electronic still camera using a memory card. It should be noted that FIG. 8 shows the case of a monochrome image for the sake of simplicity. That is, the interlaced digital image data supplied to the input terminal 11 is input to the memory controller 12. The memory controller 12 temporarily stores the input digital image data in the frame memory 13 in order to convert the digital image data into non-interlaced data. After that, the memory controller 12 reads out digital image data for each block from the frame memory 13 with 64 pixels consisting of 8 pixels in the horizontal direction × 8 pixels in the vertical direction as one block, and outputs them to the DCT circuit 14 and the encoder 15. There is.

FIG. 9 shows one block. In FIG. 9, lines 1 to 4 are the data of the first field, and lines n to (n + 3) are the data of the second field. Note that n is an arbitrary constant. Further, the pixels are read in the order of X11 to X18 of the first line, then X21 to X28 of the nth line, and finally X88 of the (n + 3) th line is read. Then, the DCT circuit 14 performs a discrete cosine transform process on the input data in block units, and outputs the transform coefficient to the encoder 15. Encoder 15
Performs the variable length coding process by performing the Huffman coding process after performing the quantization process on the input transform coefficient, and the compressed image data generated by this process is processed by the memory controller via the card controller 16. 17 is supplied.

In the case of an electronic still camera, a memory card 1
It is inconvenient for the user that it is uncertain how many still images can be recorded by 7, that is, how many shots can be taken on a normal film. Therefore, the amount of compressed image data for one still image recorded in the memory card 17 needs to be fixed. That is, no matter what subject is photographed, it is necessary to compress all digital image data corresponding to one photographed still image to a fixed data amount (code amount) or less and to fix it. .

Here, when compressing arbitrary image data to have a fixed length, how to distribute the code amount to the entire screen is an important point. For example, if the coding is performed in order from the upper part of the screen, the code amount determined in the middle may be reached, and the lower part of the screen may not be coded. For this reason, it is necessary to appropriately distribute the code amount over the entire screen for encoding so that the image quality is uniform to some extent and the image quality is not impaired.

In general, a block in which the pattern hardly changes has almost only DC (direct current) component, and conversely, a block in which the pattern changes drastically has many AC (alternating current) component in addition to the DC component. For this reason, when the compression encoding is performed, the distortion of the image quality becomes large if the AC component is deleted. Therefore, compression coding is performed so that a code amount is allocated to a block having a small frequency change amount (a small change in the pattern) and a large code amount is allocated to a block having a large frequency change amount (a large change in the pattern). By carrying out, the deterioration of the image quality is dispersed and the distortion can be made inconspicuous.

Therefore, in the encoder 15, the amount of frequency change is regarded as the amount of information, the amount of information (activity) for each block is obtained, and the amount of code that can be encoded for each block is controlled based on this activity. By doing so, fixed length is achieved. The encoder 15 adopts the two-scan method in order to realize a fixed length. In this 2-scan method, a fixed length is realized by detecting frame activity, which is the total amount of activity of one image in the first scan, and performing compression encoding in the second scan based on this frame activity. There is.

FIG. 10 shows the configuration of the encoder 15. First, the data for each block read from the frame memory 13 in the first scan is supplied to the activity calculation circuit 19 via the input terminal 18. The activity calculation circuit 19 calculates the activity Bact for each block by performing the calculation shown in the equation (1) on the input data for each block. This equation finds the sum of vertical and horizontal BPF (bandpass filter) outputs, and X (i, j) in the equation corresponds to Xij shown in FIG.

[0013]

[Equation 1]

Further, the activity calculation circuit 19 is
Activity Bac for each block calculated by equation (1)
The frame activity is calculated by adding t 1 for one frame and is output to the block bit allocation circuit 20 and the α calculation circuit 21. Of these, the α calculation circuit 21
Calculates a coefficient α for normalizing the output of the quantization table 22 based on the input frame activity.

Next, the DCT and the encoding process are performed in the second scan. That is, the output of the DCT circuit 14 is supplied to the division circuit 24 via the input terminal 23.
The value obtained by multiplying the coefficient α calculated by the α calculation circuit 21 and the output of the quantization table 22 by the multiplication circuit 25 is also supplied to the division circuit 24. Then, the division circuit 24
By dividing the output of the CT circuit 14 by the output of the multiplication circuit 25, the amplitude of the transform coefficient output from the DCT circuit 14 is controlled and supplied to the encoding circuit 26. In the encoding circuit 26, the input transform coefficient is subjected to Huffman encoding processing, and the obtained variable-length encoded data is supplied to the code amount control circuit 2
It is output to 7.

Here, the block bit allocation circuit 20
Is based on the block activity Bact calculated by the activity calculation circuit 19, the target set code amount per frame supplied via the input terminal 28, and the frame activity calculated in the first scan. The bit allocation amount of the code for each block is set and output to the code amount control circuit 27. For this reason, the code amount control circuit 27 controls the code amount of the encoded data so as not to exceed the set bit allocation amount, thereby fixing the code amount to the fixed length and outputting it to the output terminal 29.

As described above, in the conventional image compression coding apparatus as described above, since the number of quantization tables that the system can have is limited, it is set depending on the content of the image data. In the quantization table, there are cases where there is a problem in the accuracy of the code amount after the code amount control.

Further, in the above-mentioned JPEG baseline system standard, when FF (hexadecimal) appears in the byte-coded data, 00 (1
(Hexadecimal) There is a rule called byte stuffing that newly inserts data. That is, a management code (marker code) is defined in the data structure of the JPEG baseline system including encoded data,
This management code is defined by 2 bytes of FF and XX (hexadecimal, XX is a defined value other than 00). JP
The EG-compliant system manages the coded data based on this marker code. At this time, the FF, X that happens to appear in this marker code and the actual coded data.
Byte stuffing operations are performed so as not to be confused with X (hexadecimal).

For this reason, when the compression encoding processing based on the JPEG baseline method is performed while controlling the code amount, even if an attempt is made to perform encoding within the set code amount, an FF (hexadecimal) occurs due to the byte stuffing operation. Since the code amount for the amount increases, that is, data other than the originally encoded data is inserted, there is a possibility that the set code amount will be exceeded, which is a major obstacle when performing the code amount control. There are also problems.

[0020]

As described above, in the conventional image compression coding apparatus, since the number of quantization tables that the system can have is limited, the setting may be performed depending on the content of the image data. In the quantization table, the accuracy of the code amount after the code amount control may cause a problem. Also, if there is byte stuffing, data other than the original encoded data will be inserted, so there is a possibility that the set code amount will be exceeded, which is a major obstacle in controlling the code amount. I have a problem.

Therefore, the present invention has been made in view of the above circumstances, and provides an extremely good image compression coding apparatus capable of performing highly accurate code amount control without significantly increasing the scale of hardware. With the goal. Also,
It is an object of the present invention to provide a very good image compression coding apparatus capable of controlling the code amount in consideration of byte stuffing.

[0022]

An image compression coding apparatus according to the present invention comprises an orthogonal transformation means for subjecting image data to orthogonal transformation processing in block units each of which is composed of a plurality of pixels,
The transform coefficients output from the orthogonal transform means are classified into a plurality of groups on a block-by-block basis, and the transform coefficients are divided for each group by a value obtained by multiplying a quantization table prepared in advance by a scale factor, thereby performing quantization. The quantizing means for processing, the coding means for coding the transform coefficient output from the quantizing means in block units, and the code amount for each group of the coded data output from the coding means Based on the comparison result of the code amount calculating means for calculating each, and the code amount for each group calculated by the code amount calculating means and the preset code amount, the quantum is calculated from the correspondence characteristic between the code amount and the scale factor. Generating means for generating a scale factor by which the conversion table is multiplied, and the correspondence characteristic between the code amount and the scale factor based on the number of components of image data It is obtained as a control means for subjected to the generation means by shifting.

Further, the image compression coding apparatus according to the present invention comprises an orthogonal transform means for subjecting image data to an orthogonal transform process in units of blocks each consisting of a plurality of pixels, and a transform coefficient output from the orthogonal transform means. Quantization means for performing a quantization process by dividing each block into a plurality of groups and dividing a conversion coefficient for each group by a value obtained by multiplying a quantization table prepared in advance by a scale factor, and this quantization Coding means for coding the transform coefficient output from the means in block units, code amount calculation means for calculating the code amount for each group of the coded data output from the coding means, and this code The code amount for each group calculated by the amount calculation means is calculated from the byte stacking amount and the byte stack obtained from the corresponding characteristics of the scale factor. Based on the comparison result between the code amount for each group output from the adding unit and the preset code amount output from the adding device, the quantum value is calculated from the correspondence characteristic between the code amount and the scale factor. And a generation means for generating a scale factor for multiplying the conversion table.

[0024]

According to the above-mentioned structure, first, the quantization table is multiplied by the correspondence characteristic between the code amount and the scale factor based on the comparison result of the code amount for each group and the preset code amount. For the generation means for generating the scale factor, the correspondence characteristic between the code amount and the scale factor is shifted based on the number of components of the image data, so that the accuracy of the scale can be improved without significantly increasing the scale of the hardware. It becomes possible to perform high code amount control.

Further, the byte stuffing generation amount obtained from the correspondence characteristic between the byte stuffing generation amount and the scale factor is added to the code amount of each group, and the addition result is compared with a preset code amount. Based on the result, the scale factor for multiplying the quantization table is generated from the correspondence characteristic between the code amount and the scale factor, so that the code amount is prevented from exceeding the set value even if the byte stuffing process is performed. Therefore, it is possible to control the code amount in consideration of byte stuffing.

[0026]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail below with reference to the drawings. In FIG. 1, reference numeral 30 is an input terminal. The DCT conversion coefficient output from the DCT circuit 14 is supplied to the input terminal 30. That is,
As described above, the digital image data recorded in the frame memory 13 shown in FIG. 8 is scanned twice. In the first scan, the code amount calculation and the scale factor (coefficient) αt calculation described below are performed, and in the second scan, the actual encoding is performed using the calculated scale factor αt. Done.

First, the digital image data read from the frame memory 13 in the first scan is DCT.
The DCT transform coefficient is supplied to the circuit 14 to generate the DCT transform coefficient, and the DCT transform coefficient is supplied to the input terminal 30. In this case, the DCT circuit 14 classifies each block of the generated DCT transform coefficient into four groups G1, G2, G3, G4 in block units as shown in FIG.
1, G2, G3, G4 are sequentially output to the input terminal 30 for one image.

The DCT transform coefficient supplied to the input terminal 30 is supplied to the division circuit 31 and the quantization table 32.
And the scale factor output from the MPU (microprocessor unit) 33 are multiplied by a value obtained by multiplication in the multiplication circuit 34. At this time, as the scale factor supplied to the multiplication circuit 34, in synchronization with the block groups G1, G2, G3, G4 input to the input terminal 30,
Α1, α2, α3, α4 are given from the MPU 33.
That is, the group Gi (i = 1, 2, 3, 4) has scale factors αi (i = 1, 2,
3, 4) is used for quantization in the division circuit 31.

The DCT transform coefficient quantized by the division circuit 31 is supplied to the coding circuit 35 and subjected to variable length coding processing in block units, and then each group G1, G2, G.
3, code amount calculation circuits 36 and 3 installed corresponding to G4
It is supplied to 7, 38 and 39, respectively, and integrated for each group. Here, one image data is G1, G2, G
Since they are divided into four groups of G3 and G4, each integrated value output from the code amount calculation circuits 36 to 39 is about ¼ of the code integrated amount of one image data. Therefore, the code amount calculation circuits 36 to 39 multiply the integrated value by four to obtain the code amounts N1, N2, N3 and N4.

FIG. 3 shows the scale factors α1 and α.
2 shows the relationship between α2, α3 and α4 and the code amounts N1, N2, N3 and N4. Now, assuming that the set code amount Nt is at the position shown in FIG. 3, points A (α2, N2) and B (α3, N
3) The scale factor αt corresponding to the code amount Nt is calculated from the linear equation linearly approximated by the two points. Therefore, the code amounts N1, N2, N3, N4 output from the code amount calculating circuits 36, 37, 38, 39 are respectively converted into byte stuffing amount adding circuits 40, 41, 4 described later.
It is supplied to the code amount comparison circuit 44 via 2, 43, and the value closest to the set code amount Nt supplied to the input terminal 45 and the value closest to the second are selected. Then, the selected two code amounts are supplied to the αt calculation circuit 46, and the scale factor αt corresponding to the set code amount Nt is calculated from the linear equation obtained by the two-point linear approximation.

Here, paying attention to the range of the scale factor αi, when the image is encoded by this encoding algorithm, the quantized transform coefficient does not fall within the range of the preset scale factor αi (exceeds or exceeds). Below), the scale factor αt at the time of quantization
Is forcibly fixed to the maximum value or the minimum value of αi, so that the code amount after quantization may significantly exceed or fall short of the target value to be set. In order to prevent this, it is sufficient to increase the number of combinations of αi ranges set in advance, but this will lead to an increase in the hardware scale as a practical problem, so it cannot be increased unnecessarily.

Generally, in image data input for compression encoding, a parameter indicating the content of image data important at the time of compression encoding is the amount of information of the image. The fineness of the image is expressed by the DCT transform coefficient described above, and another important parameter is the number of components that make up the image. The content that can be expressed by the number of components is, for example, whether the input image is color or black and white. From the information amount of the image, the same image and the same number of pixels are included, and the monochrome image has only about half the information amount of the color image.

In this case, if the compression encoding method described above is adopted, and if the setting range of αi is assumed to be a color image, if the range is applied to a black and white image, αi will be obtained from the difference in the original information amount. Even if it is minimized, αt at the final quantization will be too large, and the final code amount will be lower than the target. Further, when the range setting of αi is performed for a black and white image, the opposite result is obtained, which exceeds the target code amount.

Therefore, in this embodiment, as the parameter for setting the range of αi, the number of input image constituent components is added in addition to the compression rate. Specifically, as shown in FIG. 4, the number of input components is 3 as a default (assuming a color image) and α1 to α4 are set. Based on this, if the number of components decreases, the left side in FIG. .Alpha.1 'to .alpha.4' are set to the right, and conversely, when the number of components is large, .alpha.1 "to .alpha.4" is set to the right in FIG. 4, and .alpha.t is determined and quantization is performed. There is.

Next, as described above, in JPEG correspondence, there is an increase in the data amount due to byte stuffing. FIG. 5 shows the relationship between the amount of byte stuffing and the scale factor α. That is, when an arbitrary image is encoded by determining the scale factor α so as to come closest to a certain set code amount (for example, 2 bit / pel), the amount of byte stuffing generated is in the lower part of the curve shown in FIG. To be distributed.

In this embodiment, the code amount calculation circuits 36-3 are used.
Byte stuffing amount adding circuits 40-4
3 is being supplied. Byte stuffing amount addition circuit 4
0 to 43, the output of each of the code amount calculation circuits 36 to 39:
By adding the byte stuffing generation amounts B1, B2, B3, and B4 according to the table shown in FIG.
Obtain N4 + B4.

The obtained code amount N1 + B1, N2 +
B2, N3 + B3, N4 + B4 are connected to the code amount comparison circuit 44.
In comparison with the set code amount Nt supplied to the input terminal 45, a value closest to the set code amount Nt and a value closest to the set code amount Nt are selected, that is, as described above, A (α2, N2 + B
2) Point 2 and point B (α3, N3 + B3) are obtained.
Then, the selected two code amounts are calculated by the αt calculation circuit 4
The scale factor αt corresponding to the set code amount Nt is calculated from the linear equation which is supplied to the No. 6 and linearly approximated to two points.

Next, in the second scan, MPU3
3, the scale factors α1, α2, α3, α4 are set to be selected, all the blocks of the image shown in FIG. 2 are sequentially input again to the input terminal 30, and they are quantized by the division circuit 31 to be encoded. After variable length coding with 35, byte stuffing circuit 47 converts into bytes and FF (hexadecimal)
When data occurs Byte stuffing process [00
(Hexadecimal insertion) is performed and the data is output from the output terminal 48.

Therefore, according to the configuration of the above-described embodiment, first, as a parameter for setting the range of the scale factor αi to be multiplied in the quantization table 32, the compression ratio as well as the number of components constituting the input image are set. If, for example, the number of components is reduced, the scale factor αi is shifted to the left in FIG. 4 by α1 ′.
.About..alpha.4 'is set, and conversely, when the number of components is large, .alpha.1 "to .alpha.4" shifted to the right in FIG. 4 are set, and .alpha.t is determined for each to perform quantization.
It becomes possible to perform highly accurate code amount control without significantly increasing the scale of hardware.

Further, by adding the byte stuffing generation amounts B1, B2, B3, B4 to the respective outputs of the code amount calculation circuits 36 to 39 according to the table shown in FIG. 5, the scale factors α1, α2, α3. Code amount N1 + B1, N2 + B2, N3 + of each group corresponding to α4
B3, N4 + B4 are obtained respectively, and from the relation between the obtained code amounts N1 + B1, N2 + B2, N3 + B3, N4 + B4 and the scale factors α1, α2, α3, α4,
Since the scale factor αt corresponding to the set code amount Nt is calculated, it is possible to prevent the code amount from exceeding the set value even if the byte stuffing process is performed, and the JPEG compliant code considering the byte stuffing. Quantity control can be performed.

Next, FIG. 6 shows another embodiment of the present invention. The same parts as those in FIG. 10 are designated by the same reference numerals. First, there is the relationship shown in FIG. 7 between the amount of byte stuffing and the frame activity. That is, when an arbitrary image is coded so as to come closest to a certain set code amount (for example, 2 bit / pel), the byte stuffing occurrence amount is distributed in the shaded portion in FIG. 7.

Therefore, in the embodiment shown in FIG. 6, the frame stuffing output from the activity calculating circuit 19 is supplied to the byte stuffing amount estimation table 49 to obtain the byte stuffing generation amount estimated from the frame activity. Then, the calculated amount of byte stuffing generated is subtracted from the subtraction circuit 5
When 0, the block bit distribution circuit 20 is controlled by a value subtracted from the set code amount supplied to the input terminal 28, thereby performing the code amount control in consideration of byte stuffing.

As another example, as the block activity, a DCT transform coefficient may be input to the input terminal 18 and the sum of absolute values of its AC components may be taken.
The present invention is not limited to the above-described embodiments, but can be variously modified and implemented without departing from the scope of the invention.

[0044]

As described above in detail, according to the present invention,
It is possible to provide an extremely good image compression encoding device capable of performing highly accurate code amount control without significantly increasing the scale of hardware. According to the invention,
It is possible to provide an extremely good image compression coding apparatus capable of controlling the code amount in consideration of byte stuffing.

[Brief description of drawings]

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

FIG. 2 is a diagram for explaining grouping of DCT transform coefficients.

FIG. 3 is a diagram showing a relationship between a code amount and a scale factor.

FIG. 4 is a diagram shown for explaining a main part of the embodiment.

FIG. 5 is a diagram showing a relationship between a byte stuffing occurrence amount and a scale factor.

FIG. 6 is a block diagram showing another embodiment of the present invention.

FIG. 7 is a diagram showing a relationship between a byte stuffing occurrence amount and a frame activity.

FIG. 8 is a block configuration diagram showing a conventional image compression encoding device.

FIG. 9 is a diagram showing one block configuration of 8 pixels in the horizontal direction × 8 pixels in the vertical direction.

FIG. 10 is a block diagram showing an encoder of the conventional device.

[Explanation of symbols]

11 ... Input terminal, 12 ... Memory controller, 13 ... Frame memory, 14 ... DCT circuit, 15 ... Encoder,
16 ... Card controller, 17 ... Memory card, 18
... input terminal, 19 ... activity calculation circuit, 20 ... block bit allocation circuit, 21 ... α calculation circuit, 22 ... quantization table, 23 ... input terminal, 24 ... division circuit, 25 ...
Multiplier circuit, 26 ... Encoding circuit, 27 ... Code amount control circuit,
28 ... input terminal, 29 ... output terminal, 30 ... input terminal, 3
1 ... Division circuit, 32 ... Quantization table, 33 ... MPU,
34 ... Multiplier circuit, 35 ... Encoding circuit, 36-39 ... Code amount calculating circuit, 40-43 ... Byte stuffing amount adding circuit, 44 ... Code amount comparing circuit, 45 ... Input terminal, 46 ...
αt calculation circuit, 47 ... Byte stuffing circuit, 48
... output terminal, 49 ... byte stuffing amount estimation table, 50 ... subtraction circuit.

Claims (2)

[Claims] 1. An orthogonal transformation means for subjecting image data to orthogonal transformation processing in block units each of which is composed of a plurality of pixels,
The transform coefficients output from the orthogonal transform means are classified into a plurality of groups on a block-by-block basis, and the transform coefficients for each group are divided by a value obtained by multiplying a quantization table prepared in advance by a scale factor, thereby performing quantization. The quantizing means for processing, the coding means for coding the transform coefficient output from the quantizing means in block units, and the code amount for each group of the coded data output from the coding means Based on the comparison result between the code amount calculation means for calculating each, and the code amount for each group calculated by the code amount calculation means and the preset code amount, from the corresponding characteristics of the code amount and the scale factor, Generating means for generating the scale factor by which the quantization table is multiplied, and the code amount and the scale factor based on the number of components of the image data. Data and the corresponding characteristic picture compression encoding apparatus characterized by comprising by shifting and a control means for subjected to the generating means. 2. An orthogonal transform means for subjecting image data to orthogonal transform processing in block units each of which is composed of a plurality of pixels,
The transform coefficients output from the orthogonal transform means are classified into a plurality of groups on a block-by-block basis, and the transform coefficients for each group are divided by a value obtained by multiplying a quantization table prepared in advance by a scale factor, thereby performing quantization. The quantizing means for processing, the coding means for coding the transform coefficient output from the quantizing means in block units, and the code amount for each group of the coded data output from the coding means A code amount calculating means for calculating each and a code amount for each group calculated by the code amount calculating means, and an adding means for adding the byte stuffing generation amount obtained from the correspondence characteristic between the byte stuffing generation amount and the scale factor, respectively. Based on the result of comparison between the code amount for each group output from the adding means and the preset code amount. Image compression encoding apparatus characterized by the corresponding characteristics of the quantity and the scale factor becomes comprises a generating means for generating said scaling factor to be multiplied with the quantization table.