JPH0775398B2 - Image processing system - Google Patents

Description

Description: TECHNICAL FIELD The present invention compresses image data having gradations for each pixel block to reduce the data amount of image information and restores compressed image data. Image processing system.

(Prior Art) Generally, in an image processing apparatus such as a plate-making scanner, by reading an original image, image data is obtained for each pixel which is a fixed minute area. The amount of image data in the printing field is larger than that of television image data, and the amount of information per image is several MB to several tens of MB. To store such a large amount of data as a database as it is, a huge memory is required, and the cost of data transmission becomes large.

In order to deal with this, an encoding technique for reducing the information amount of an image, that is, an image data compression technique is generally used.

(Problems to be Solved by the Invention) However, when the original image is compressed at a high data compression rate to obtain compressed data, the image quality of the restored image obtained by restoring the compressed image deteriorates as compared with the original image. Therefore, if data compression / decompression is repeated a plurality of times, there is a problem that the image quality gradually deteriorates.

(Object of the Invention) The present invention has been made in order to solve the above-mentioned problems in the prior art, and is capable of suppressing the progress of image quality deterioration when data compression / decompression is repeated a plurality of times. The purpose is to provide.

(Means for Achieving the Purpose) In order to achieve the above object, according to the present invention, compressed image data is obtained by an image data compression apparatus, and the compressed image data is restored by an image data restoration apparatus to restore restored image data. In the image processing system, (a) the image data compression device (a-1) statistic for each of a plurality of pixel blocks into which an image is divided, for obtaining a statistic representing a distribution state of image data in the pixel block. Amount calculation means, (a-2) determination means for determining whether the image data is restored image data, and (a-3) when the image data is restored image data, the restored image According to the first procedure identification data given to the data, one of the plurality of different compression procedures is selected, while the image data is restored image data. If it is not, then one of the plurality of compression procedures is selected according to the statistic to specify one specific compression procedure for compressing the image data for each pixel block. And (a-4) a second procedure for identifying the specific compression procedure by calculating the compressed image data by compressing the image data for each pixel block according to the specific compression procedure. Compression means for storing or transmitting with the identification data.

Further, (b) the image data restoration device is (b-1)
The transmitted or stored compressed image data and the second
Reading means for reading the procedure identification data of (b-
2) Decompression means for decompressing the compressed image data and generating decompressed image data by a decompression procedure according to the specific compression procedure indicated by the read second procedure identification data, and (b-3) ) A compression procedure designation recording unit that generates data instructing the specific compression procedure as the first procedure identification data and includes it in the restored image data.

Further, the first procedure identification data may be expressed by using the numerical value of the least significant digit of the restored image data at a predetermined pixel position in the pixel block.

(Operation) By including the procedure identification data in the restored image data restored from the compressed data, the same compression procedure as when the original compressed image data was obtained can be selected when compressing again. Even when the compression / decompression is repeated, the deterioration of the image quality is suppressed.

If this procedure identification data is represented by the least significant digit of the image data at a predetermined pixel position in the pixel block, it is not necessary to store only the procedure identification data independently, and the data of the restored image data The amount can be reduced.

(Example) Here, A. a method for creating compressed image data, B. a method for creating decompressed image data, C. a schematic configuration and operation of the device, D. an example of data compression / decompression, and E. a modified example, Will be sequentially described.

A. Method of Creating Compressed Image Data FIG. 12 is a conceptual diagram showing an example of image blocking in the image processing system according to the embodiment of the present invention. The image 1 is composed of (X × Y) pixels P, and is divided into (V × H) pixels by a pixel block B _vh having a plurality of pixels. FIG. 13 is a conceptual diagram showing the configuration of one pixel block B _vh . The pixel block B _vh is composed of (M × N) pixels p _ij , and image data f is calculated for each pixel.
_ij is obtained. In the example of FIG. 13, the pixel block B _vh
Is composed of (8 × 8) pixels, but may be composed of (4 × 4) pixels or (16 × 16) pixels.

In data compression, first, pixel block B in image 1
Orthogonal transformation of image data is performed for each _vh . As the orthogonal transform, various methods such as discrete Fourier transform and Hadamard transform can be used, but here, discrete cosine transform (hereinafter referred to as “DCT transform”) represented by the following equation is used.

Further, m and n usually take values in the range of 0 to M-1, 0 to N-1, respectively.

FIG. 14 shows image data f in one pixel block.
FIG. 6 is an explanatory diagram showing _ij and the conversion coefficient F _mn obtained by the equation (1). Hereinafter, the table showing the conversion coefficient F _mn is referred to as “conversion coefficient table FT”. The conversion coefficient F ₀₀ is the image data f _ij
Of the image data (hereinafter referred to as “DC component”), and other conversion factors F _mn (hereinafter referred to as “AC component”) indicate the distribution state of the image data, each of which is represented by, for example, an 8-bit binary number. This is the represented value. However, in FIG. 14, it is shown in decimal notation for convenience.

As can be seen from the image data f _{ij in} FIGS. 14 (a-1) and (b-1) and the corresponding conversion coefficient F _mn in FIGS. 14 (a-2) and (b-2). , The position (m, where the conversion coefficient F _mn has a non-zero value depending on the distribution of the image data f _ij
n), and the magnitude of the values are quite different. Here, the conversion coefficient F _mn , whose absolute value is large, is a component that well represents the characteristics of the distribution state of the original image data f _ij , and the number of conversion coefficients F _{mn with} a large absolute value is the image data f _ij. The larger the spatial change of, the greater the tendency is to increase.

In this embodiment, paying attention to the bias of the distribution of the conversion coefficient F _mn as described above, a code table that specifies only a part of the AC component as valid is used. FIG. 15 is an explanatory diagram showing an example of a code table. The coordinate (m, n) to which the number of effective bits I _mn is given in the code table CT indicates that the conversion coefficient F _mn at the corresponding position (m, n) is designated as effective. Further, each effective bit number I _mn is the maximum number of bits when representing the conversion coefficient F _mn . That is, the maximum effective value F
The values of the conversion coefficient F _mn up to _max (= (2 I _mn power) -1) are valid as they are. In the figure, "-" indicates that the number of valid bits is "0". The number of effective bits I _mn may all be the same value (for example, 8 bits), but it is smaller than the number of bits (8 bits) of the original conversion coefficient F _mn , and is appropriate for each coordinate (m, n). If such a value (for example, 6 bits, 3 bits, etc.) is used, there is an advantage that the compression rate increases when the compressed image data is obtained later. In addition, when the value of the conversion coefficient F _mn is larger than the maximum effective value F _max ,
The conversion coefficient F _mn is assumed to be equal to the maximum effective value F _max .
The number of effective bits I _mn is empirically determined so that such a case does not occur, and even if the conversion coefficient F _mn is equal to the maximum effective value F _max , its influence can be ignored.

A plurality of code tables CT that are different from each other in terms of the position of the effective bit number I _mn and the value pattern are prepared in advance. The pattern of the effective bit number I _mn of these code tables CT is empirically required so that a higher quality image can be reproduced with a smaller number of conversion coefficients. For example, it is set based on the pattern having a high appearance frequency as the above-mentioned pattern from among the characteristic patterns of general patterns such as a person, a landscape, and a still life.

Next, with respect to the conversion coefficient table FT obtained for each pixel block B _vh , the optimum code table is selected from the plurality of code tables CT. For this selection, the sum of effective conversion coefficients (hereinafter, simply referred to as "effective sum") _SF is obtained by the following formula.

Then, the effective sum _SF is obtained for each code table CT, and the code table CT having the largest value is selected as the optimum code table.

The conversion coefficient F _mn that is valid in each code table CT
Number (hereinafter, "effective AC frequency components" as referred.) If the N _c the same, selection of the optimal code table according to the effective sum S _F described above, becomes more reliable.

The compressed image data includes the average value of the original image data f _ij , the table number of the selected optimum code table CT, and the conversion coefficient F _mn designated as valid in the code table CT.
It is created based on the value of and. The number of effective AC components N _c is the total number of conversion coefficients F _mn (M ×
N), and much less than the individual transform coefficients F _mn
Since the value of is represented by the smaller number of effective bits I _mn , the compression rate becomes extremely large. The compression rate can be further increased by encoding the compressed image data with a known Huffman code or the like.

Next, a method of dividing the code table CT into a plurality of groups is adopted in order to further increase the compression rate according to the characteristics of the image. For this purpose, first, the standard deviation σ of the image data f _ij is _calculated for each pixel block B _vh . And standard deviation σ
The code table is divided into a plurality of compression groups (code table groups) according to the size of

Compressed group g ₁ : CT _{11 to} CT _1L when σ <σ ₁ Compressed group g ₂ : CT when σ ₁ ≦ σ <σ _{2 21 to} CT _2L CT when compressed group g ₃ : σ ₂ ≦ σ <σ _{3 31 to} CT _3L compression group g ₄ : When σ ₃ ≦ σ, CT _{41 to} CT _4L (where σ ₁ , σ ₂ and σ ₃ are predetermined threshold values) of the code table of each compression group g _{1 to} g ₄ . Number L
May have different values.

FIG. 16 is a diagram showing an example of a distribution G of the number A of appearances in the entire image 1 with respect to the standard deviation σ obtained for each pixel block B _vh of a certain image 1. In the figure, the number of appearances A on the vertical axis
Are shown on a logarithmic scale.

The first threshold σ ₁ is defined as a value at which the change in the image within the pixel block is small when the standard deviation σ is smaller than this. Since the value of the AC component of the conversion coefficient of such a pixel block is small, a small number of effective AC components N _c
The image quality is not deteriorated by using the code table having Further, as can be seen from FIG. 16, since the appearance rate A of the pixel block with a generally small standard deviation σ is high,
All these pixel blocks have a small number of effective AC components N _c
If it is represented by the code table of, the effect of increasing the compression rate is great.

When the standard deviation σ is between the first threshold value σ ₁ and the second threshold value σ ₂ , it indicates that the image is gradually changing within the pixel block. In this case, in order to maintain good image quality, the number of effective AC components larger than that of compression group g ₁
It is better to use a code table with N _c . Since this compression group g ₂ also has a relatively high appearance rate A, the effect of improving the compression rate is also high.

For the standard deviation σ of the second threshold σ ₂ or more, the appearance rate A
Are relatively low, but these are all large effective AC component numbers
If it is processed by the code table of N _c , it will cause a reduction in the compression rate. Therefore, the third threshold σ ₃ is set as an intermediate value.

Code tables CT _{1k to} CT _4k (k for each compression group g _{1 to} g ₄
= 1 to L), the number of effective AC components N _{c1 to} N _c4 has the following relationship.

0 ≦ N _c1 ≦ N _c2 ≦ N _c3 ≦ N _c4 ≦ M × N−1 (3) In each compression group, the number of effective AC components in the code table is constant. Further, when N _c1 = 0, the compressed image data of the pixel block B _vh that is determined to belong to the compression group g ₁ is represented by only the average value of the pixel data f _ij .

In data compression, you prepare in advance the threshold value σ ₁ ~σ ₃ and the compression group g ₁ to g ₄ grouped code table groups CT _1k to CT _4k to the above. Then, based on the value of the standard deviation σ of the pixel block B _vh , the compression group g
Any one of _{1 to} g ₄ is selected, and the optimum code table is selected from the code table group belonging to the one compression group.

B. Method of Constructing _Decompressed Image Data _Decompressed image data f _ij of a certain pixel block is obtained by _performing DCT inverse transform on the transform coefficient F _mn in the compressed image data.

The inverse DCT transform is given by: At the time of this restoration, information indicating which of the compression groups g _{1 to} g ₄ was used to create the original compressed image data is recorded in the restored image data. Hereinafter, such information will be referred to as “compressed group information”.

In addition, in order to record the compression group information in the restored image data f _ij , it is necessary to record the same information in the compressed image data. In the compressed image data, the code table number has a role of recording the compression group information, as described later.

The compression group information in the restored image data f _ij is incorporated in a part of the image data by using the least significant digit of the image data at a predetermined pixel position in the pixel block, for example.

When the image data is represented by binary data, the two compression groups can be distinguished by whether the least significant bit of the image data is “0” or “1”. Below, setting the least significant bit to “0” is simply referred to as “evenization”, and setting to “1” is referred to as “oddization”.

Since the compression group is divided into four now, the compression group information can be represented by 2 bits. Therefore, two pieces of image data f ₁₁ and f ₃₃ are selected in the pixel block B _vh shown in FIG. 13, the values are adjusted as follows, and the compression group information is recorded: compression group g ₁ : f Both ₁₁ and f ₃₃ are even compression group g ₂ : f ₁₁ is odd, f ₃₃ is even compression group g ₃ : f ₁₁ is even, f ₃₃ is odd compression group g ₄ : f ₁₁ and f ₃₃ Both are odd-numbered. Even-numbering of the image data f ₁₁ is performed, for example, according to the following formula: f ₁₁ = ff ₁₁ & 254 (5) where ff ₁₁ : image data before even-numbering f ₁₁ : image data after even-numbering &: Bit logical product operator The bit logical product operator & represents an operation of taking a logical product for each bit when data is represented by a binary number.
If (5) expressed by an 8-bit binary number the equation becomes the following _{equation: e 7 e 6 e 5 e} 4 e 3 e 2 e 1 e 0 = a 7 a 6 a 5 a 4 a 3 a 2 a 1 a ₀ & 11111110 (6) where e _i (i = 0 to 7): each bit of the image data f ₁₁ a _i (i = 0 to 7): each bit of the image data ff ₁₁ bit AND operator & From the function of, the equation (6) can be rewritten as the following equation.

e ₇ e ₆ e ₅ e ₄ e ₃ e ₂ e ₁ e ₀ = a ₇ a ₆ a ₅ a ₄ a ₃ a ₂ a ₁ 0 (7) That is, what number is the original image data ff _11? However, the least significant bit e ₀ of the image data f ₁₁ converted by the equation (5) is always 0. Therefore, the even numbering of the image data f ₁₁ is achieved by the equation (5). Moreover, since the difference between the image data f ₁₁ after the evenization and the original image data ff ₁₁ is 1 or less, the difference on the image quality is negligible.

On the other hand, the oddification of the image data is performed by the following formula: f ₁₁ = (ff ₁₁ & 254) +1 (8) Since the least significant bit of the image data f _{11 on} the left side is always 1, the image data f ₁₁ is It is an odd number. Moreover, the first on the right side
Since the maximum value of the term (f ₁₁ & 254) is 254, the image data f
The maximum value of ₁₁ is 255. Therefore, the maximum value 255 of the image data will not be exceeded by the odd numbering according to the equation (8). Further, since the difference between the odd-numbered image data f ₁₁ and the original image data ff ₁₁ is 1 or less, the influence on the image quality is negligible.

The above has described the evenization and the oddization of the image data f ₁₁ , but the same applies to the image data f ₃₃ .

Based on the values of the least significant bits e ₁₁₀ and e ₃₃₀ of the evenized or oddized image data f ₁₁ and f ₃₃ , the compression group information N _g is represented by the following 2-bit binary number. : N _g = e ₃₃₀ e ₁₁₀ (9) The relationship between the compression group information N _g and the compression groups g _{1 to} g ₄ is as follows.

Compression group g ₁ : N _g ＝ 00… (10a) Compression group g ₂ : N _g ＝ 01… （10b） Compression group g ₃ : N _g ＝ 10… （10c） Compression group g ₄ : N _g ＝ 11… ( 10d) Note that each bit forming the compressed group information N _g is not limited to the least significant bit of the image data f ₁₁ and f ₃₃ , and the least significant bit of the image data at any two pixel positions can be used. Further, since the number of compression groups is 4 in this embodiment, the compression group information N _g is represented by a 2-bit binary number. However, when the number of compression groups is larger, the number of bits may be increased according to the number. Good.

As described above, if the restored image data is configured to include the compression group information, when the restored image data is compressed again, it can be treated as belonging to the same compression group as before.

C. General Configuration and Operation of Device FIG. 1 is a schematic configuration diagram showing an image data compression device for compressing image data as one component of the image processing system of one embodiment of the present invention. In the figure, an image data compression device AP includes an image memory 5, a pixel block readout controller 6, a buffer memory 7, a two-dimensional DCT converter 8, a linear quantizer 9, an encoder 10, and an encoded data recording controller. 11, a compression group information read controller 12, a standard deviation calculator 14, a compression group information controller 15, a ROM 17, and a controller 13. The standard deviation calculator 14 corresponds to the statistic calculating means in this invention, the compression group information read controller 12 corresponds to the determining means in this invention, and the compression group information controller 15 is the specific procedure designating means in this invention. The encoder 10 and the encoded data recording controller 11 correspond to the compression means in the present invention. The operation of the image data compression device AP will be described below with reference to the flowchart shown in FIG.

In step S1, the standard deviation thresholds σ _{1 to} σ ₃ are input from the parameter input terminal 16 provided outside the data compression apparatus AP and set in the compression group information controller 15. Steps S2 to S4 and S11 to S13 are steps of sequentially selecting pixel blocks B _vh to be processed.

In step S5, the image data f _ij in one pixel block B _vh is read from the image memory 5 by the pixel block read controller 6 and stored in the buffer memory 7.

Step S6 is a step of determining the compression group information N _g . Here, first, in step S61, it is determined whether the image data f _ij is a restored image.
In the case of the image data of the restored image, the data indicating the restored image may be written in the header portion of the data file of the image data, and the above determination may be made depending on the presence or absence of the data.

When the image data f _ij is not the restored image but the original image data, steps S62 and S63 are executed. First, in step S62, the image data f _ij is _supplied from the buffer memory 7 to the standard deviation calculator 14 to calculate the standard deviation (statistical amount) σ. In step S63, the standard deviation σ is given from the calculator 14 to the compression group information controller 15 to determine the compression group information N _g . That is, the compression group information controller 15 compares the standard deviation σ with the thresholds σ _{1 to} σ ₃ to determine which of the compression groups g _{1 to} g ₄ the pixel block belongs to. And compression group information
N _g is determined as in equations (10a) to (10d).

On the other hand, when the image data f _ij is the restored image, step S64 is executed. In this step,
As shown in the equation (9), image data of two pixel positions
The compression group information N _g is combined by the least significant bits (first procedure identification data) e ₁₁₀ and e _{330 of} f ₁₁ and f ₃₃ . According to FIG. 1, the image data f ₁₁ and f ₃₃ are compressed group information read controller from the buffer memory 7.
Read out to 12. The read controller 12 synthesizes the compressed group information N _g from these image data f ₁₁ and f _33, and supplies this to the compressed group information controller 15. Thus, when the compression group information N _g is given from the read controller 12, the controller 15 holds the compression group information N _g, determined compression group information N _g from the standard deviation σ as described above No action is taken.

When the compression group information N _g is determined in step S6, the image data f _ij is transferred from the buffer memory 7 to 2 in step S7.
It is transmitted to the three-dimensional DCT converter 8 for DCT conversion.

In step S8, the transform coefficient F _mn is linearly quantized. Linear quantization is, as is generally known, a transformation coefficient F _mn
Is divided by a predetermined value α (“quantum width”), and the value F
_{When mn} / α contains a fractional part, it means to convert it to an integer. The value of the quantum width α may be fixed, or may change according to the standard deviation σ. Note that linear quantization may not be performed in some cases, as will be described later.

In step S9, the optimum code table is selected for the transform coefficient table FT ^* having the linearly quantized transform coefficient _Fmn ^* .

FIG. 3 is a flowchart showing details of the optimum code table selection process performed by the encoder 10 in step S9. The meaning of each symbol in the figure is as follows.

max: effective sum S parameter for determining the maximum value of _F Max: Parameter indicates the number of the code table is the maximum effective sum S _F is l: Parameter indicates the number of the code table L: number of code tables c: effective AC Parameter N _c indicating the component number: number of effective AC components First, in step S21, which of the compression groups g _{1 to} g ₄ the target pixel block corresponds to is compressed group information.
It is identified based on N _g . Steps S22 to S30 are the effective sum S of the L code tables of the compression group.
_This is the procedure for selecting the one that maximizes _F (see equation (2)).

In steps S23 to S26 for obtaining the effective sum S _F , the sub table T as shown in FIG. 4 stored in the ROM 17 is used. The sub-table T has one compression group g _i (i =
Code table CT _{g1 to} CT _gL in any one of 1 to 4)
For each of the above, only the coordinate value (m _lc , n _lc ) of the coordinate position where the value of the effective bit number I _mn is not “0” is shown. Here, 1 = 0 to (L-1) and c = 0 to ( _Nc- 1).
Further, the subscript g of the code tables CT _{g1 to} CT _gL indicates the compression group g _i . For one code table CT _gl , the coordinates (m _lc , n _lc ) are called one after another from the sub-table T, and among the conversion coefficients F _mn ^* of the conversion coefficient table FT ^* ,
Only absolute values of those at the above coordinates are added in step S24. The code table with the maximum effective sum _SF is selected as the optimum code table. When the optimum code table is selected, the conversion coefficient of the conversion coefficient table FT ^*
Of F _mn ^* , only the effective AC component specified by the optimum code table is the effective bit number I _mn (eg 6 bits)
, The conversion factor other than the effective AC component F _mn ^*
Is considered to be none. In step S31, the table number Max of the optimum code table and the conversion coefficient F _mn ^* designated as valid in the optimum code table are converted into a Huffman code string, and compressed image data is obtained (steps S31 and S31).
Ten). Note that the average value of the original image data f _ij may be Huffman-encoded at the same time at this time, but separately from step S31, only the average value may be encoded by a method such as so-called predictive encoding.

The compressed image data (encoded data) D _f thus obtained
Is transmitted from the encoder 10 to the encoded data recording controller 11 and further output to the image memory 5 or the transmission path 2 to an external circuit (not shown). This output selection command is input to the parameter input terminal 16 by the operator prior to the compression of the image data, and is initialized in the encoded data recording controller 11.

FIG. 5 is a conceptual diagram showing the file structure of the encoded compressed image data D _f . The header 18 in FIG. 5A includes thresholds σ ₁ , σ ₂ , σ _{3 and the} like, and the code string of each pixel block is sequentially arranged following the header 18. Further, FIG. 5B shows the internal structure of the code string 19 of each pixel block, and the table number Ma of the optimum code table is shown.
x (second procedure identification data) and a code string of the conversion coefficient F _mn ^* validated in the optimum code table.

FIG. 6 is a schematic configuration diagram showing an image data decoding device for decoding encoded compressed image data D _f as one component of the image processing system of the embodiment of the present invention. The image data decoding device IAP includes an image memory 20, a read controller 22, a buffer memory 23, a decoder 24, a ROM 25, an inverse quantizer 26, a two-dimensional DCT inverse converter 27, a compression group information recorder 28,
And a recording controller 29. The read controller 22 corresponds to the read means in the present invention, the decoder 24, the inverse quantizer 26 and the two-dimensional DCT inverse converter 27 correspond to the restore means in the present invention, and the compression group information recorder. 28 corresponds to the compression procedure designation recording means in this invention.

The compressed image data D _f is read from the image memory 20 or the transmission path 21 by the read controller 22, given to the buffer memory 23, and further transmitted to the decoder 24. ROM2
In 5, the same code table used for encoding the image data, the Huffman code table for decoding the Huffman code string, etc. are recorded. The compressed image data D _f is decoded by the decoder 24 into transform coefficients F _mn ^* . Conversion factor F
_{The mn} ^* is inversely quantized by the inverse quantizer 26. That is, the transform coefficient F _mn ^* (= F _mn / α) linearly quantized with the quantum width α is multiplied by the quantum width α. The conversion factor thus obtained
F _mn is inversely transformed by the two-dimensional DCT inverse transformer 27 to obtain image data ff _ij for each pixel in the pixel block.

Based on the image data ff _ij thus obtained, the compression group information recorder 28 records the compression group information in the image data ff _ij . FIG. 7 is a flowchart showing the recording procedure. In step S31, the compression group information recorder 28 reads the code table number Max in the compressed image data D _f from the buffer memory 23. In step S32, the compression group information N _g is determined from the code table number Max (second procedure identification data). And
By steps S33a to S33c and steps S34a to S34d, the image data ff ₁₁ (f ₁₁ ) and ff ₃₃ (f ₃₃ ) are made odd and even. Steps S34a to S34d are the equations (5) described above.
And (8).

The image data f _ij in which the compressed group information N _g is recorded is recorded in the image memory 20 by the recording controller 29. If the above decoding process is performed for all pixel blocks, the image data f _{ij of the} entire image can be obtained.

When the image data f _ij once restored is compressed again, the compression group information N _g is read from the least significant bit (first procedure identification data) of the two image data f ₁₁ and f ₃₃ . Then, data compression is performed using the compression group identified by the compression group information N _g .

D. Example of data compression / decompression Example 1: When linear quantization of F _mn is not performed. FIG. 8 shows image data f _ij and conversion when image data compression / decompression is repeated by applying this embodiment. Coefficient F _mn
FIG. In the figure, image data f
_ij (0) represents the image data of the original image, and f _ij (k) (k =
1 to 3) show image data obtained by the k-th decompression after alternately receiving data compression and decompression. Also, F
_mn (k) (k = 1 to 3) indicates a conversion coefficient obtained by the k-th compression after alternately receiving data compression and decompression.

In this example, the linear quantization of the transform coefficient F _mn is not performed. Therefore, the conversion factor F _{mn of the decimal} value obtained by the DCT transformation
Is stored as the compressed image data D _t without being converted into an integer and is restored. However, in the figure, for simplicity, only the integer part of the conversion coefficient F _mn is displayed. On the other hand, the image data f _ij (k) is always saved and restored as an integer value.

First, the image data f _ij (0) of the original image is DCT-transformed to obtain a transform coefficient _mn ⁰ (1). This conversion coefficient _mn ⁰ (1) does not have the selection of the effective AC component using the code table, and has 19 AC components.

Then, as a result of selecting only the effective AC component using the code table, the conversion coefficient F _mn (1) is obtained. The conversion coefficient F _mn (1) has six effective AC components F ₀₁ , F ₀₂ , F ₁₀ , F ₁₁ , F ₁₃ , and F ₁₄ .

Furthermore, the restored image data f _ij (1) is obtained by performing the DCT inverse transform (IDCT) on the transform coefficient F _mn (1).

In the example of FIG. 8, the standard deviation threshold value σ ₁ is 4.0, σ ₂
And the value of σ ₃ are set to 5.0 or more. Since the standard deviation σ of the original image data f _ij (0) is 4.862 as shown in the figure, this image data f _ij (0) belongs to the compression group g ₂ .

On the other hand, the image data f _ij restored after one-time data compression
Since the standard deviation σ of (1) is 3.621, if classification is performed by the value of standard deviation σ, the image data f _ij (1) is compressed group g.
_Will belong to ₁ . But, as I already explained,
Information indicating that the image data f _ij (0) of the original image belongs to the compression group g ₂ , that is, compression group information N _g =
01 (binary number display) is recorded by the even / odd of data f ₁₁ (1) and f ₃₃ (1) (indicated by underline in the figure) in the image data f _ij (1). In particular,
As described in Section B, the data f ₁₁ (1) is an odd number, f ₃₃ (1)
Has been adjusted to an even number. Then, the second and subsequent data compressions are executed assuming that the image data f _ij (1) belongs to the compression group g ₂ .

In this example, since the transform coefficient F _mn (k) is stored as a decimal value with high precision, the DCT transform is performed reversibly.
Therefore, the results of the second and subsequent data compression and decompression are equal to the first results.

FIG. 9 shows the transition of the image quality deterioration of the restored image as the image data f.
It is expressed as the S / N ratio of _ij (k). The S / N ratio R _SN corresponding to the case (example 1) shown in FIG. 8 is indicated by a white circle. S / N ratio R _SN ratio R _SN is defined by the following formula: R _SN = 10log ₁₀ (255 ² / r)… (11a) Where Δf _ij is the difference between the original image data and the restored image data. In addition, X and Y in the equation (11b) respectively indicate the number of columns and the number of rows of all pixels existing in the image (see FIG. 12).

As can be seen from FIG. 9, in Example 1, the restored image data f
_{After ij} (1), the effect that the S / N ratio R _SN does not change and the deterioration of image quality does not progress appears.

FIG. 9 also shows, for comparison, an example in which data compression and decompression are repeated without using the present invention.
In this method of identifying the compression group using the value of the standard deviation σ, it is determined that the compression group belongs to the compression group g ₁ during the second data compression. Therefore, the image quality of the restored image after being subjected to the second data compression is significantly deteriorated,
The value of the S / N ratio R _SN also drops significantly as shown in FIG.

Examples 2, 3: Case of performing linear quantization of F _mn Example 1 above is a case of not performing linear quantization of the transform coefficient F _mn . However, by performing the linear quantization of transform coefficients F _mn, the storage capacity required for the storage of the transform coefficients F _mn can be greatly reduced, there is an advantage that it is possible to further increase the data compression ratio.

FIG. 10 shows the image data f _ij (k) and the conversion coefficient F _{mn in the same} manner as in FIG. 8 when the conversion coefficient F _mn is linearly quantized.
It is explanatory drawing which shows ^* (k). Subscript * for conversion factor
The reason why the symbol is added is to show that it is a linearly quantized one.

Since the value of the quantum width α is “1”, the “linear quantization” here is equivalent to the “integer ratio”. Moreover, this integer conversion is performed by rounding off.

S / N of image data f _ij (k) in the case shown in FIG. 10 (example 2)
The ratio R _SN is shown in FIG. 9 as a black square. As can be seen from the figure, when F _mn is linearly quantized (Example 2), S /
Although the rate of decrease in the N-ratio R _SN is large, the degree of decrease in the S / N-ratio R _SN is small compared to the comparative example, and it can be said that the image quality deterioration is also small.

As described above, the deterioration of the image quality in the example 2 is different from the case of the example 1, and an error caused by rounding off the decimal point when linearly quantizing the conversion coefficient F _mn (integralization here) is accumulated. It will be done.

In order to further suppress the deterioration of image quality, the following method can be adopted.

Before performing DCT inverse conversion based on the conversion coefficient F _mn ^{* (k),} to modify the value of the effective AC is the component transform coefficient F _mn ^{* (k)} as follows: F _mn ^*> 0 FF when _mn = _Fmn ^* + 0.5 (12a) When _Fmn ^* <0, _FFmn = _Fmn ^* -0.5 (12b) Then, the DCT inverse transform is performed using the modified transform coefficient _FFmn . Hereinafter, applying equations (12a) and (12b) will be referred to as “correction of conversion coefficient F _mn ^* ”.

An example (example 3) of image data f _ij (k) and conversion coefficient F _mn ^* when data compression and decompression are performed by this method is given _below .
It is shown in Figure 11. In addition, the S / N ratio R _{SN of} this Example 3 is 9th.
It is represented by a white square in the figure. As can be seen from FIG. 9, in Example 3, the decrease in the S / N ratio R _SN is small, and the S / N ratio R _SN is kept constant in the second and subsequent data compression and decompression.

Equations (12a) and (12b) are introduced based on the following ideas. That is, as shown in FIG. 10 (example 2), the image data f _ij (1) restored without modifying the conversion coefficient F _mn ^* and the image data f _ij (0) of the original image are By comparison, the restored image data f _ij (1) has a smaller difference in data for each pixel. That is, the image data that has been once compressed and decompressed becomes more averaged data. On the other hand, transform coefficients F _mn ^* of the absolute value | F _mn ^{* |,} as shown in the equation (4) is the amplitude of each frequency component of the image data f _ij to be restored on the basis of the transformation coefficients F _mn ^* . Therefore, if the absolute value | F _mn ^* | is increased, the amplitude of each frequency component is increased, and the change in the restored image data f _ij within the pixel block is increased. (1
Equations 2a) and (12b) are equivalent to increasing the absolute value | F _mn ^* | in this way. Therefore, (12a), (12
If the conversion coefficient F _mn ^* is corrected by the expression b) and the image data is restored, the effect of averaging the data by the data compression can be canceled by the correction by these expressions.

The expressions (12a) and (12b) are generally expressed as follows.

When F _mn ^* > 0 FF _mn = F _mn ^* + α (13a) When F _mn ^* <0 FF _mn = F _mn ^* -β (13b) Here, α and β are both positive values , Especially 0.3-0.5
It is preferably in the range of.

In this way, if the transform coefficient F _mn ^* is corrected and restored, as shown in FIG. 9 as Example 3, the S / N ratio R _SN that is almost equal to that when linear quantization is not performed (Example 1) is maintained. Therefore, the effect of suppressing the deterioration of image quality is great. Moreover, the conversion factor
Since F _mn ^* is linearly quantized and stored, there is an advantage that the size of the compressed image data D _t can be reduced and a high data compression rate can be achieved.

E. Modifications In the above embodiment, the code table is classified into a plurality of code table groups according to the standard deviation of the image data in the pixel block, but not limited to the standard deviation, the variance and various similar statistics are used. You may classify. That is,
The classification may be performed using some statistic that statistically indicates the distribution state of the image data.

Further, in the above embodiment, the method of compressing the image data of one image has been described, but it goes without saying that the present invention may be applied to each color-separated image for a color image. For example, YMCK print image, RGB
Image data can be compressed for each color-separated image for signal images and Y _S · I _S · Q _S signal (chroma signal) images. In this case, for each color-separated image, the number of standard deviation thresholds and their values, the number of code tables in each code table group, the number of effective AC components of the code table, the pattern of the number of effective bits of the code table, etc. By changing, it is possible to perform compression suitable for each color separation image.

At this time, the pixel position (i, j) ((1,1) and (3,3) in the above embodiment) for recording the compressed group information in the image data f _ij is set for each color separation image. It is preferable to determine the positions so that they are different from each other. This is because when the compression loop information is recorded, the value of the least significant digit of the image data f _ij at these pixel positions is adjusted, so that the change caused by the adjustment causes a slight change in image quality, and therefore each color is changed. This is because, in the case of creating a reproduced image by synthesizing decomposed images, if the pixel positions of which the values of the least significant digits are not overlapped with each other, the change in image quality is reduced as a whole. From the same meaning, it is preferable that these pixel positions are not adjacent to other pixel positions in each color separation image.

In the above embodiment, the compression group information is expressed and recorded using the numerical value of the least significant digit of the image data f _ij of the restored image, but the compression group information is recorded as data different from the image data f _ij. You may do it. In this case, the image data is simply data representing the density value at each pixel position (hereinafter referred to as “image density data”) f _ij .
It is defined as a broad concept that includes not only this but also other data representing compressed group information.

The method of data compression and decompression is not limited to that of the above embodiment, and various methods such as the method disclosed in Japanese Patent Laid-Open No. 55-109086 can be applied. The present invention is applicable if a method is adopted in which a statistic indicating the distribution state of image data in a pixel block is obtained and one of a plurality of different compression procedures is selected according to the statistic. It is possible.

In the above embodiment, performing data compression using different code table groups for the respective compression groups g _{1 to} g ₄ corresponds to the “plurality of compression procedures”.

When a plurality of compression procedures are not distinguished in the form of compression groups as in the above-described embodiment, it is sufficient that the restored image data includes some procedure identification data for identifying the selected compression procedure. At this time, the compressed image data is also saved together with such procedure identification data.
In the above embodiment, it goes without saying that the image data f ₁₁ , f ₃₃ in the restored image data f _ij and the code table number Max in the compressed image data D _t respectively represent the procedure identification data. .

(Effects of the Invention) As described above, according to the present invention, since the restored image data is created so as to include the procedure identification data, when the restored image data is compressed again, it is used for the previous data compression. Since the same compression procedure can be applied, it is possible to suppress the progress of image quality deterioration when data compression and decompression are repeated a plurality of times.

Further, if the procedure identification data is expressed by using the numerical value of the least significant digit of the image data at a predetermined pixel position in the pixel block, it is not necessary to store the procedure identification data independently,
There is an effect that the data amount of the restored image data can be reduced while minimizing the deterioration of the image quality of the restored image.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic configuration diagram of a compression device to which an embodiment of the present invention is applied, FIGS. 2 and 3 are flowcharts showing an embodiment of the present invention, and FIG. Is a conceptual diagram showing an example of a sub-table, FIG. 5 is a conceptual diagram showing a configuration of compressed image data, FIG. 6 is a schematic configuration diagram showing a decoding device, and FIG. 7 is a compression group information. Flowchart showing the procedure of the recording method, FIGS. 8, 10 and 11 are explanatory diagrams showing examples of data compression and decompression, and FIG. 9 is a diagram showing the S / N ratio of the decompressed image, FIG. Fig. 13 is an explanatory diagram showing image blocking, Fig. 13 is an explanatory diagram showing a pixel array in a pixel block, Fig. 14 is an explanatory diagram showing an example of image data and transform coefficients, Fig. 15 is a code. Fig. 16 is a conceptual diagram showing an example of a table, and Fig. 16 shows an example of the relationship between the standard deviation of image data and the number of appearances. It is an explanatory diagram. AP ... image data compression device, IAP ... image data decoding device, B _vh ... pixel block, CT ... code table, I _mn ... effective number of bits, F _mn ... conversion coefficient, f _ij ... image Data, D _f ... compressed image data, N _g ... compressed group information, P ... pixels, σ ... standard deviation

Claims (2)

[Claims] 1. An image processing system in which compressed image data is obtained by an image data compression device, and the compressed image data is restored by an image data restoration device to obtain restored image data, wherein: (a) the image data compression device (A-1) For each of a plurality of pixel blocks into which the image is divided, a statistic calculating unit that obtains a statistic representing the distribution state of the image data in the pixel block, and (a-2) the image data is a restored image. (A-3) When the image data is restored image data, the determination means determines whether or not the data is different from each other according to the first procedure identification data given to the restored image data. While selecting one of the plurality of compression procedures, when the image data is not the restored image data, the plurality of compression procedures are performed according to the statistic. Specific procedure designating means for designating one specific compression procedure for compressing the image data for each pixel block by selecting one of the procedures; and (a-4) the specific procedure. Compression means for compressing the image data for each pixel block according to a compression procedure to obtain compressed image data, and storing or transmitting the compressed image data together with second procedure identification data for identifying the specific compression procedure. (B) The image data decompression device includes: (b-1) a reading unit that reads the transmitted or saved compressed image data and the second procedure identification data; and (b-2) the read unit. Decompression means for decompressing the compressed image data and generating decompressed image data by a decompression procedure according to the specific compression procedure instructed by the second procedure identification data, (b-3) Serial image processing system characterized by comprising: a compression procedure specified recording unit operable to generate data indicating a particular compression procedure included in the restored image data as the first procedure identification data. 2. The first procedure identification data is expressed by using the numerical value of the least significant digit of the restored image data at a predetermined pixel position in a pixel block.
The image processing system described.