JP2003264705A - Image encoder and image encoding method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To form images having frequency characteristics based on the distribution of quantization data by image kinds and excellent reproducibility. <P>SOLUTION: The multilevel image data of a prescribed block size are decomposed into orthogonal transformation coefficients, an amplitude distribution for each space frequency is detected, classification into image modes is performed and they are converted to the intrinsic number of bits for each image mode. Then, the orthogonal transformation coefficients are weighted corresponding to the space frequency, valid coefficients are selected preferentially in the descending order of a weighted amplitude absolute value from the transformation coefficients after quantization for each block and remaining transformation coefficients are invalidated. Then, an encoding part 6 allocates a quantized prescribed bit length code and a valid flag to the valid coefficient, allocates only an invalid flag indicating that it is invalid to the invalid coefficient and performs encoding. Also, feedback is performed to the selection of the valid coefficients so as to limit a code amount for each block to be within a fixed amount. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding processing apparatus for compressing and restoring multi-valued image data, and in particular, it is capable of gradationally recording multi-valued image data having continuous gradation read by a scanner or the like. The present invention relates to an image encoding device and an image encoding method applicable to a copying machine having a printer.

[0002]

2. Description of the Related Art In a conventional digital multi-function peripheral, image data read from a scanner unit is stored in an image memory, edited after being rotated or combined, and then read from the image memory and printed from a printer unit. For example, when a printer feeds recording paper vertically on a document placed horizontally on the reading surface of the scanner unit, the image data that has been read is rotated to match the orientation of the recording paper. Was possible.

On the other hand, a multi-value printer capable of recording not only binary data but also halftone data has been developed. Since the multi-valued printer can express the image data, which has been represented by 1 bit per pixel so far, by the multi-valued data of 2 bits or more, a continuous tone image can be recorded and the image quality can be improved.

When handling multi-valued image data, the amount of data increases several times as compared with binary data, so that the storage capacity of the image memory for accumulating image data also increases, leading to an increase in cost. Although the memory capacity can be suppressed by storing the image data after compressing it, there is a problem that the compressed image data cannot be edited depending on the compression method. For example, JPE, which is a widely used compression method
Since G coding is variable length coding, it is difficult to edit because pixel position information cannot be stored. Therefore, JPE
Editing must be performed before G encoding, but even with this method, the cost of the editing memory increases because the capacity of the editing memory increases. In addition, there is a problem that the processing speed decreases unless the memory access speed is increased due to the large amount of edit data.

To solve such a problem, as a compression / decompression device for a multi-value gradation data image, orthogonal transform coding is known as an effective means for lossy coding which realizes code data reduction with a small distortion. . The image is two-dimensionally orthogonally transformed for each block, and the obtained transform coefficients are quantized with different quantization bit numbers. In order not to exceed a certain amount in a page or in a block, encoding has been aborted just before the code amount exceeds a certain amount. Generally, human visual characteristics, that is, it is sensitive to low frequency components in spatial frequency, but it is not noticeable even if there are some errors in high frequency components Quantization and encoding were performed, and the encoding was discontinued so as not to exceed a certain limit amount.

However, according to this method, when the energy distribution is concentrated in a high spatial frequency, if the truncation occurs before the encoding of the high frequency component, the spatial frequency component having a large energy distribution is lost. Therefore, there is a problem in image reproducibility.

The present inventor has previously proposed a method of allocating a significant number of quantization bits in the order of energy of spatial frequencies, that is, in order of increasing absolute value of transform coefficient, and adjusting the code amount in a block. This coding method is a method of selecting in the order of absolute values of transform coefficients at all spatial frequencies, coding those within a range that does not exceed the code amount limit value within a block as significant coefficients, and truncating the others. is there.

The proposed method will be described below with reference to the drawings.

FIG. 13 is a block diagram of a conventional image coding apparatus. Such an image encoding device includes an image input device 11
Configures a two-dimensional space having a block size by controlling the writing of raster image data obtained by scanning an original into the line memory 12. Then, the orthogonal transformation unit 13 orthogonally transforms the two-dimensional image in order of each block size. While the image type determination unit 18 determines the image type from the energy distribution of the orthogonal transformation coefficient, the quantization unit 14 quantizes the orthogonal quantization coefficient with a bit length corresponding to the image type. The selecting unit 15 is configured such that the orthogonal transform coefficients are selected in the order of absolute values within a range that does not exceed a predetermined code amount, and the quantized coefficient after selection is encoded by the encoding unit 16.

In the image coding apparatus configured as described above, the coefficient selected as the effective coefficient in the selection unit 15 is classified into the image type determined as the block and the unique spatial frequency of the image type. The result quantized with the quantized bit length is used as it is, while the insignificant coefficient is regarded as 0 and encoded. That is, while the entire block is fixed length coding, the variable length coding is performed for each coefficient.

As described above, by adjusting the number of significant coefficients and the number of non-significant coefficients and the bit allocation, it is possible to limit the code amount of each block within a certain amount, and to minimize the image deterioration after compression / decompression. You can keep it to the limit.

[0012]

However, in the above image coding apparatus, the energies of the transform coefficients of all spatial frequencies are compared equally, and the spatial frequency characteristic and the human visual characteristic peculiar to each image type are compared. Considering the above, there are cases in which the method is not necessarily the optimum method in terms of the balance between image reproducibility and coding efficiency. That is, in the conventional method, if the high-frequency component that is difficult to be recognized has a higher energy than the low-frequency component, it is selected and encoded, but this does not contribute much to the reproducibility of a general image, or rather, it is low frequency. It is often better to enrich the components because of noise generation and natural reproducibility. An image in which the density is expressed by high frequency modulation, such as a halftone image, is better reproduced by cutting the high frequency component sharply.

As described above, depending on the type of image, it may be necessary to set different spatial frequency characteristics to be reproduced. Further, it may be necessary to correct the frequency characteristic with respect to the characteristic of the combined printer or the image processing characteristic of the subsequent stage.

The present invention has been made in view of the above-mentioned circumstances, and it is possible to form an image having excellent reproducibility because each image type can have a frequency characteristic by distribution of quantized data. Another object of the present invention is to provide a simple image coding apparatus and image coding method.

[0015]

According to the present invention, encoding is performed for each spatial frequency in each image type based on image type information determined from the distribution of orthogonal transform coefficients or image type information obtained from the outside. The weighting is applied to the coefficient absolute value that is referred to when selecting the significant transform coefficient to be targeted, and the distribution of the code data is biased to give the frequency characteristic in the multi-value compression encoding.

[0016]

BEST MODE FOR CARRYING OUT THE INVENTION A first aspect of the present invention is an orthogonal transform means for decomposing multi-valued image data of a predetermined block size into orthogonal transform coefficients corresponding to spatial frequencies, and an amplitude of the orthogonal transform coefficients for each spatial frequency. Image classification means for detecting a distribution and classifying into image modes, quantizing means for converting the orthogonal transform coefficient into the number of bits peculiar to each spatial frequency for each image mode, and the orthogonal transform coefficient according to the spatial frequency. Weighting is performed, and for each block, the effective coefficient is preferentially selected from the quantized conversion coefficient in descending order of the amplitude absolute value of the weighted conversion coefficient, and the remaining conversion coefficient is invalidated, and effective coefficient. A variable length encoding means for allocating the quantized predetermined bit length code and a valid flag to the invalid coefficient, and allocating only the invalid flag to the invalid coefficient. Is the amount of code image encoding apparatus comprising a code amount limiting means for performing feedback to the selection means so as to limit within a certain.

According to this configuration, since the conversion coefficient having large energy is selected and the code amount in one block is kept within a fixed amount, it is possible to achieve both the code efficiency and the image reproducibility. Since the weighting of the absolute value of the transform coefficient can be set for each spatial frequency domain, the frequency characteristic of the image after compression conversion can be adjusted.

According to a second aspect of the present invention, in the image coding apparatus according to the first aspect, a weighting coefficient is set according to a combination of each spatial frequency and each image mode, and the weighting coefficient is used. The orthogonal transform coefficients are weighted.

As a result, the frequency characteristic can be adjusted according to the characteristic peculiar to the image type.

According to a third aspect of the present invention, in the image coding apparatus according to the first and second aspects, the image classification means detects the amplitude distribution of the spatial frequency of the orthogonal transform coefficient and classifies the image mode. In addition, it was decided to capture the external image type determination results and use them for image mode classification.

Thus, the code whose frequency characteristic is adjusted for a wider range of image types is used by using the discrimination result of the external image discriminating means for the image type which cannot be discriminated only by the energy distribution of the conversion coefficient. Can deal with

According to a fourth aspect of the present invention, in the image coding apparatus according to the first, second and third aspects, the coding means selectively applies a variable length coding format and a fixed length coding format. It was decided to.

With this configuration, the coding efficiency can be increased when the number of target transform coefficients is small, such as when limiting the frequency distribution of transform coefficients.

According to a fifth aspect of the present invention, multivalued image data of a predetermined block size is decomposed into orthogonal transform coefficients corresponding to spatial frequencies, the amplitude distribution of the orthogonal transform coefficients for each spatial frequency is detected, and classified into image modes. Then, the orthogonal transform coefficient is converted into the number of bits unique to each spatial frequency for each image mode, the orthogonal transform coefficient is weighted according to the spatial frequency, and the block is weighted from the quantized transform coefficient for each block. The effective coefficient is preferentially selected in the descending order of the absolute amplitude value of the converted coefficient, the remaining conversion coefficient is invalidated, the quantized predetermined bit length code and the valid flag are assigned to the effective coefficient, and the invalid coefficient is set as the invalid coefficient. Performs variable length coding that assigns only an invalid flag indicating that it is invalid, and feeds back to the selection of effective coefficients so that the code amount for each block is limited within a certain amount. An image coding method comprising and.

An embodiment of an image coding apparatus / method of the present invention will be specifically described below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of an image coding apparatus according to an embodiment of the present invention. As shown in the figure, the image input device 1 supplies digital data to the image encoding device. The image input device 1 outputs digital 8-bit digital image data by raster scanning.

The digital image data supplied from the image input device 1 is stored in the line memory 2. In the present embodiment, the line memory 2 is used to accumulate digital image data for each line for eight lines to generate block image data for 8 × 8 pixels.

The image data converted into a block of 8 × 8 pixels by the line memory 2 is input to the orthogonal transformation unit 3. The orthogonal transform unit 3 performs Hadamard transform of an 8 × 8 matrix in this example. FIG. 5 shows a basis image of the Walsh-Hadamard transform as an example of the orthogonal transform. FIG. 6A shows image data (X00 to X7) of 8 × 8 pixels before orthogonal transformation.
7), and FIG. 6B shows the transform coefficients after orthogonal transform arranged in the base arrangement of FIG. The orthogonally transformed transform coefficient is used by the quantization unit 4 and the image type determination unit 8
Are supplied in parallel.

The quantizer 4 quantizes the transform coefficient input from the orthogonal transformer 3 with a predetermined number of bits.

The image type discriminating unit 8 monitors the energy distribution state of the conversion coefficient and discriminates, for each block, a photographic image area, a line image area, or an intermediate image area based on the bias and intensity.

The weighting calculation unit 9 has a weighting coefficient LUT.
It is composed of a (look-up table) and outputs weighting coefficients at different spatial frequencies for each image type.

The quantized data output from the quantization unit 4 and the weighting coefficient output from the weighting calculation unit 9 are input to the selection unit 5. In the selection unit 5, the absolute amplitude value of the coefficient is multiplied and the weighting coefficient is multiplied, the calculation results are arranged in the order of absolute values, the code amount necessary for each encoding is integrated, and when the limit amount is reached, subsequent conversion is performed. The coefficient is invalidated and output to the subsequent stage.

An encoding unit 6 is arranged at the output stage of the selecting unit 5. The encoding unit 6 has a code length defined based on whether the coefficient determined by the selection unit 5 is invalid or valid,
Encoding is performed in the code scan order. Code data memory 7
Accumulates the data encoded by the encoding unit 6.

Next, the operation of this embodiment configured as described above will be described. 2 and 3 are flowcharts showing a series of processes from the input of a pixel block to the completion of encoding. The digital image data input to this apparatus is converted into a block of 8 × 8 pixels by the line memory 2 (step S1), and the image data is orthogonally transformed by the orthogonal transformation unit 3 in units of the block (step S).
2).

In the image type discriminating section 8, the conversion coefficients Y00 to Y00
The coefficient which is not 0 is counted from Y77 (FIG. 6B) and the count value (N) of the non-zero coefficient in one block is stored (step S3). Then, the count value (N) is compared with the threshold value α (step S4). If N> α, the process branches to step S5, and otherwise, the process branches to step S7.

If the process branches to step S7, it means that the block is non-zero, that is, the number of effective transform coefficients is small in the block. Therefore, in FIG. 6B, it is determined that the region is a so-called photographic image region where the AC component distribution is small. it can. In this case, the parameter setting giving priority to gradation is performed in step S7. That is, since there are few effective coefficients in the quantized transform coefficient and there is a margin in the code amount, the number of bits (code length) for quantizing the effective coefficient, for example, the code length is defined as LL, LM, and LH for each frequency domain. In this case, as shown in FIG. 4, the set values are the numerical values LPL, LPM and LPH that are increased from the normal values LML, LMM and LMH. When the number of bits for quantizing the effective coefficient is increased, the size of the quantization step becomes a relatively small numerical value, so that the quantization accuracy is increased and the gradation reproduction in the photographic area is prioritized. In addition, the block distortion caused by the quantization error is reduced.

On the other hand, if the process branches to step S5, nonzero conversion coefficients are counted from the high frequency region as shown in FIG. 7A and the count value M is stored. Then, the count value (M) is compared with the threshold value β (step S
6) If M> β, the process branches to step S8; otherwise, the process branches to step S9.

In the case of branching to step S8, it can be determined that it is a line drawing area because there are non-zero coefficients in the high frequency area. In this case, the parameters are set with priority on the resolution. That is, the number of quantization bits (code length) is set to a value smaller than the normal value LDL, LDM, LDH in order to assign the transform coefficients existing in all frequency regions from the low frequency region to the high frequency region to the codes as effective coefficients. Set to. Since the code length of the effective coefficient becomes small, more effective coefficients are assigned to the code. As a result, it is possible to prevent a problem in which significant coefficients in a high frequency region, which are later in the encoding order, are lost due to the termination of encoding, and the reproducibility of an image having a high spatial frequency such as a line drawing is improved.

When the process branches to step S9, it can be determined that the image is an intermediate image between the photographic area and the line drawing area. In such a case, normal parameters are set.
In the present embodiment, the code length normal values LML, LMM, L
It is set to MH. The image type determination unit 8 performs step S
The image type result md determined by any one of 7, 8, and 9 is output to the quantization unit 4, the weighting calculation unit 9, and the encoding unit 6.

Further, in the present embodiment, a special image type determination result such as a halftone dot image area detection result detected by an external spatial frequency detecting means (not shown) such as a Fourier transform means is used to determine the image type. It is input to the unit 8 as an external determination result.

The image type determination unit 8 determines whether or not a halftone image is detected based on the external determination result (step S1).
0), if the external determination result indicates halftone dot image detection, the image type md is rewritten to the halftone dot image type (step S1).
1).

The quantizing unit 4 acquires the code length corresponding to the image type result md from the table of FIG. 4, and sets the acquired code length as the code length setting values LL, LM, LH. Then, the transform coefficients Y00 to Y77 are quantized by the number of bits indicated by the code length setting values LL, LM, and LH, and the quantized coefficients are stored in the coefficient buffer memory (not shown) in the selection unit 5 (step). S12). FIG. 10 shows a configuration example of the coefficient buffer memory. FIG. 10A shows a state before sorting, which will be described later, and the coefficient after quantization is written as the “quantization coefficient”. The post-calculation coefficient is a calculation result by the weighting calculation unit 9 described later.

The weighting calculation unit 9 refers to the weighting factors WL, WM and WH for the image type result md by the look-up table shown in FIG. In the look-up table, as shown in FIG. 8, the weighting factors WL, WM, and WH are defined in a register for each of the low frequency region, the intermediate frequency region, and the high frequency region of the spatial frequency corresponding to the image type. Has been done. As shown in the table, when the image type is a photographic image, WPL and W
Defined by PM and WPH, and similarly WD for line image type
L, WDM, WDH, and WML, W for intermediate image types
It becomes MM and WMH. By allowing weighting to be set arbitrarily for these image types, it is possible to set the bias for each frequency of the conversion base to be selected, and the spatial frequency characteristics of the image after compression / expansion can be changed with a certain degree of freedom. The weighting coefficient corresponding to the orthogonal transformation coefficient is multiplied and output to the selection unit 5 as a "post-calculation coefficient", and when the image type is not a halftone image (step S1).
3) stored in the coefficient buffer memory (not shown) in association with the “quantized coefficient” (step S1)
4).

The selecting section 5 reads the weighted coefficients stored in the coefficient buffer memory and sorts them so that they are arranged in descending order of the absolute amplitude values. As shown in FIG. It is stored together with the quantized coefficient in a sort table (not shown) (step S14). The encoding target on the sort memory is i = 1 (step S15), and i
The th coefficient is added to the encoding target (step S17).
Further, the code amount of the quantized coefficient is counted in descending order from the sort table, and if the limit code amount is reached (step S18), the subsequent coefficient is set to 0 in the coefficient buffer memory.
The value is cleared to be an invalid coefficient, and the processing of the selection unit 5 ends (step 21). In addition, it is necessary to take into consideration not only the simple code length of each individual transform coefficient but also the total number of flags included in the code when integrating the code amount.

Here, the code amount is a combination of valid / invalid flags and valid data arranged in code order. When i selected codes and the maximum code rank among them are j, the code amount BL of one block is BL = M + D + Σ + j, where M is the coding type code, D is the DC base code, and Σ
Is the total code amount of i selected valid data codes.

When this is compared with the code limit amount of one block as a whole (step S18) and the limit amount is not exceeded (step S19) and all the coefficients have not been encoded, the next rank in the sort memory is determined. The coefficient is selected as an encoding target (step S20) and added to the encoding target (step S17). If the limit amount is exceeded in step S18, 0 is cleared in the coefficient buffer for the i-th and subsequent coefficients in the sort order (step 21). If the total code amount does not reach the limited code amount in step S18,
If the code rank i does not reach the number of codes in step S19, the code rank is incremented (step S20), and the process returns to step S17 to continue the in-block processing.

As a result, even if the transform coefficients positioned after the code rank have a large amplitude absolute value after the weighting calculation, these are preferentially coded as effective coefficients. Therefore, it is possible to prevent the loss of the component to be prioritized.

The coding unit 6 reads the coefficients from the coefficient buffer memory of the selection unit 5 in code order, which is the scan order of coefficients, and executes the coding process for each valid coefficient and invalid coefficient (step S22). Specifically, if the coefficient read from the coefficient buffer memory is a valid coefficient that is a non-zero coefficient, a flag "1" indicating that the coefficient is a valid coefficient and "valid data" that is the numerical value of the valid coefficient are set. Output the concatenated code string. If the quantized coefficient read from the coefficient buffer memory is an invalid coefficient of 0, only the flag "0" indicating the invalid coefficient is encoded.

In step S23, it is monitored whether or not the coding of all the coefficients in the block is completed, and step S22 is repeated until the coding is completed. When the encoding of one block is completed, the block code shown in FIG. 11A is obtained, and this is stored in the code data memory 7.

FIG. 9 shows an example of the structure of a block code for one block coded by the coding unit 6. FIG. 9 (A)
As shown in, the block code of one block is composed of an encoding type code and code data. The encoding type code is, as shown in FIG.
“Image type” in which the image type md obtained in
The code lengths LL, LM, and LH obtained by the image type discrimination unit 8 are described, and "effective data length" is described. As shown in FIG. 9C, the code data portion is composed of DC component and AC component codes. In the figure, the code data per block is 56 bits, and the DC component code is 8 bits. It shows an example in which 48 bits are used for the code of the AC component. FIG. 9D shows that when the coefficient code of the AC component is an invalid coefficient, it is composed of only the flag "0", and when it is a valid coefficient, it is composed of a flag "1" and 3 bits of valid data.

FIG. 11 shows an example of coding for one block. FIG. 11A shows a one-dimensional bit map.
(B) shows an example of arrangement in a two-dimensional bitmap.

When the coding in the block is completed, the process proceeds to step S26, it is judged whether or not the coding of all the images is completed, and if the images to be coded remain, the step S26 is executed.
Returning to 1, the encoding of the next pixel block is started.

Further, in the present embodiment, so-called variable length coding in which the effective coefficient is coded with a specified quantization length, but the invalid coefficient is coded with only the shortest flag, and all coefficients are coded with a fixed length. The so-called fixed-length coding is appropriately switched and applied.

Usually, in coding, variable length coding is used for each coefficient, but when the number of coefficients is small to limit the frequency domain and all the coefficients in the domain are coded without omission. For a fixed length, the efficiency is improved because the overhead for the flag is eliminated. For example, suppression of moire and roughness is often more important than reproducibility of high frequency components such as a halftone image. These image deterioration factors are often noise components such as mosquito noise resulting from the lack of low frequency components,
Therefore, it is effective to exclude all high-frequency components and encode the bases of all low and intermediate frequencies.

Then, in step S13 of FIG. 3, it is judged whether the type is a halftone image type, and if not, the flow is from step S13 described above. If it is a halftone image type, the process proceeds to step S24, and the quantized code from the coefficient buffer is sequentially encoded as it is with a fixed length (step S24). This is continued until the coefficient to be coded limited within one block ends (step S2).
5). Then, when all the images are completed, the processing is completed (step S26).

An example of this encoding is shown in FIG. 12
In (A), like the variable-length coding, the coding type code 8 bits and the DC code 8 bits as the image type are followed by the codes VL0 to VL2 of the transform coefficients in the low frequency region, and the code length LL (4 bits in this case). ), And the codes VM0 to VM11 of the transform coefficients in the intermediate frequency domain are quantized and arranged with the code length LM (in this case, 3 bits). FIG. 12B shows an example of arrangement in a two-dimensional bitmap.

As described above, according to the present embodiment, since a conversion coefficient having a large energy is selected and the code amount in one block is kept within a fixed amount, both code efficiency and image reproducibility can be achieved. In addition, since the weighting of the absolute value of the transform coefficient can be set for each spatial frequency domain, the frequency characteristic of the image after the compression transform can be adjusted.

Further, according to the present embodiment, since the conversion coefficient weighting setting for each spatial frequency domain can be set for each image type, the frequency characteristic can be adjusted according to the characteristic peculiar to the image type.

Further, according to the present embodiment, since the external image type discrimination result can be taken in, the image type which cannot be discriminated by only the energy distribution of the conversion coefficient is discriminated by the external image discriminating means. By using, it is possible to cope with the coding in which the frequency characteristic is adjusted for a wider range of image types.

In addition, since it is possible to select all transform coefficients to be encoded within a limited range in advance and add a coding method having the same fixed length, it is possible to limit the frequency distribution of transform coefficients. When the number of target transform coefficients is small, the coding efficiency can be increased.

[0061]

As is apparent from the above description, according to the present invention, since the frequency of the conversion base can be weighted for each image type, the spatial frequency characteristic of the reproduced image after compression / expansion can be manipulated. It is possible to adjust to a natural and reproducible setting for each image type.

Further, according to the present invention, variable length coding or fixed length coding is selected for each transform coefficient depending on the number of coefficients to be coded, and the coding efficiency is higher and the image reproducibility is higher. It is possible to realize good multi-tone encoding / decoding.

Further, according to the present invention, since the image type can be switched based on the identification information from the outside, it is possible to perform appropriate code distribution based on a more advanced discrimination result.

Further, according to the present invention, so-called variable length coding in which effective coefficients are coded with a specified quantization length, but invalid coefficients are coded only with the shortest flag, and all coefficients are coded in fixed length. Since so-called fixed length coding is appropriately switched and applied, coding efficiency can be improved when the number of target transform coefficients is small, such as when limiting the frequency distribution of transform coefficients.

[Brief description of drawings]

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

FIG. 2 is an operation flowchart of the first half of the image encoding device according to the above embodiment.

FIG. 3 is an operation flowchart of the latter half of the image encoding device according to the above embodiment.

FIG. 4 is a diagram showing an image type-quantization length table of the image encoding device according to the embodiment.

FIG. 5 is a diagram showing a conversion base of Hadamard transform in the above embodiment.

FIG. 6A is a diagram showing pixel values of an input image block before orthogonal transform, and FIG. 6B is a diagram showing transform coefficients after orthogonal transform.

FIG. 7A is a diagram showing an example of frequency domain definition of transform coefficients, and FIG. 7B is a diagram showing code scan order of transform coefficients.

FIG. 8 is a diagram showing an image type-weighting coefficient table of the image encoding device according to the embodiment.

9A is an overall view of a block code of a block encoded in the above embodiment, FIG. 9B is a configuration diagram of an encoding type code, FIG. 9C is a configuration diagram of code data, and FIG. 9D is a conversion coefficient of an AC component. Diagram showing an example of encoding

FIG. 10A is a diagram showing an example of a transform coefficient buffer. (B) Diagram showing sort table

FIG. 11A illustrates an example of a one-dimensional bitmap code when a coefficient variable length code is used. FIG. 11B illustrates a two-dimensional bitmap code example when a coefficient variable length code.

FIG. 12A illustrates a one-dimensional bitmap code example when a coefficient fixed-length code is used, and FIG. 12B illustrates a two-dimensional bitmap code example when a coefficient fixed-length code is used.

FIG. 13 is a block diagram showing the configuration of a conventional image encoding device.

[Explanation of symbols]

1 Image input device 2 line memory 3 Orthogonal transformation unit 4 Quantizer 5 Selector 6 Encoding section 7 Code data memory 8 Image type discrimination unit 9 Weighting calculation section

   ─────────────────────────────────────────────────── ─── Continued front page    F term (reference) 5C059 KK22 MA21 MC11 MC38 ME01                       PP01 SS28 TA41 TA48 TB13                       TC04 TC24 TD10 TD12 UA02                 5C078 AA04 BA53 CA22 CA25 CA31                       DA01

Claims (5)

[Claims] 1. An orthogonal transform means for decomposing multi-valued image data of a predetermined block size into orthogonal transform coefficients corresponding to spatial frequencies, and an image classification for detecting an amplitude distribution of the orthogonal transform coefficients for each spatial frequency and classifying into image modes. Means, a quantizing means for converting the orthogonal transform coefficient into a number of bits peculiar to each spatial frequency for each image mode, and weighting the orthogonal transform coefficient according to the spatial frequency, and after quantizing for each block Selection coefficient for prioritizing the effective coefficients in descending order of the amplitude absolute value of the weighted conversion coefficient and invalidating the remaining conversion coefficient, and the effective coefficient is the quantized predetermined bit length code. And a valid flag are assigned to the invalid coefficient, and only an invalid flag indicating that the invalid coefficient is invalid is assigned to the variable length format encoding means. An image coding apparatus comprising a code amount limiting means for feeding back to the selecting means. 2. The image code according to claim 1, wherein a weighting coefficient is set according to a combination of each spatial frequency and each image mode, and the orthogonal transform coefficient is weighted using the weighting coefficient. Device. 3. The image classification means detects the amplitude distribution of the spatial frequency of the orthogonal transform coefficient and classifies it into an image mode, and fetches an external image type determination result to use for classification of the image mode. The image coding apparatus according to claim 1 or 2, characterized in that. 4. The image encoding method according to claim 1, wherein the encoding means selectively applies a variable length encoding format and a fixed length encoding format. apparatus. 5. The multi-valued image data of a predetermined block size is decomposed into orthogonal transform coefficients corresponding to spatial frequencies, the amplitude distribution of the orthogonal transform coefficients for each spatial frequency is detected, and the orthogonal transform coefficients are classified into image modes. The number of bits unique to each spatial frequency is converted for each image mode, the orthogonal transform coefficient is weighted according to the spatial frequency, and the absolute amplitude of the transform coefficient weighted from the quantized transform coefficient for each block. The effective coefficient is selected preferentially in descending order of value, the remaining transform coefficients are made invalid, the quantized predetermined bit length code and the effective flag are assigned to the effective coefficient, and it is shown that the invalid coefficient is invalid. An image code characterized by performing variable-length coding in which only invalid flags are assigned and performing feedback to select effective coefficients so as to limit the code amount of each block within a certain range. Encoding method.