JP2000316174A - Color picture compression method and its device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently encode and compress color picture data without deteriorating the picture quality. SOLUTION: Color picture data are divided into plural pixel blocks, and stored in each G, R, and B color 4×4 buffer 501, 511, and 521, and two hierarchical sub-band conversion is operated for each color by sub-band converting parts 502, 512, and 522. The G color picture is encoded independently as specific components, and R and B color low frequency coefficients are independently encoded. A G color high frequency coefficient and R and B color high frequency coefficient vectors are prepared for R and B color high frequency coefficients, and inter-high frequency coefficient vector rate is calculated, and information indicating the rate is quantized and encoded. Then, a difference value between the R and B color high frequency coefficients decoded from the information indicating the quantized rate and the actual R and B color high frequency coefficients is encoded.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a color image compression method and a color image compression apparatus, and more particularly, to a color image compression method and a color image compression apparatus for compressing color image data composed of a plurality of components. About.

[0002]

2. Description of the Related Art Conventionally, the demand for color images has been increasing year by year, and the resolution has been increased to meet the demand for high image quality. For this reason, in order to reduce the amount of memory, for example, in an image processing device used for a digital copying machine, a facsimile device, a digital camera, a digital video, etc., or an image recording device such as a CD-ROM, a floppy disk, etc. There is a growing demand for efficient compression of data.

A color image is generally composed of RGB (red, green,
Blue) and CMYK (cyan, magenta, yellow,
It is represented by superimposing several colors such as four colors of black), but these colors do not have independent values and are correlated with each other. It is also common practice to decompose these colors into luminance information and color difference information, but there is also a strong correlation between the luminance and the color difference. The correlation between the luminance and the color difference is mainly related to a fine structure of an image such as a local inclination.

The compression of color image data by the conventional color image compression method has been performed by separately compressing each color or luminance / color difference. However, if the compression is performed using the above-described correlation, the redundancy can be further reduced. I understand. For example, according to JP-A-5-300383,
Color image information is divided into luminance and chrominance, and encoding is performed based on the ratio of the AC component of chrominance to the AC component of luminance.

[0005]

However, in such a conventional color image compression method, for example, as disclosed in the above-mentioned Japanese Patent Application Laid-Open No. Hei 5-300383, an AC component of a color difference and an AC component of luminance are used. Since the amplitude ratio was directly encoded, the improvement in compression efficiency was still insufficient, and both AC components (hereinafter referred to as high-frequency coefficients) were forcibly applied in an area where the luminance and the color difference were not in a proportional relationship (such as a color halftone area). ), There is a problem that the image quality deteriorates.

The present invention has been made in view of the above problems, and provides a color image compression method and a color image compression apparatus capable of efficiently compressing color image data without deteriorating image quality. The purpose is to:

[0007]

In order to achieve the above object, a color image compression method according to claim 1 is a color image compression method for compressing color image data composed of a plurality of components. Dividing the color image data into n × m (n and m are natural numbers) pixel blocks, and converting the information of each component into a representative value and a coefficient representing a fine structure for each pixel block. When compressing the color image data, a representative value of all components, a coefficient representing a fine structure of only a specific component, a coefficient representing a fine structure of a component other than the specific component, and a fine The encoding is performed using the coefficient representing the relationship with the coefficient representing the structure.

The "representative value" is obtained by substituting all the components in a pixel block with their densities, such as the average value of each component or the average value of luminance / color difference information. Means a density that does not change the rough impression of the image.

The "coefficient representing the fine structure" is information representing fine information (inclination, etc.) of the image in the pixel block, and the original image is obtained by combining with the "representative value" described above. It means a coefficient that can be completely or almost completely restored.

[0010] The above "component" refers to R
Elements that constitute each color such as GB or color image data such as luminance and color difference components. In addition, the above “specific component” means that one of the components of the color image data or the color
It means one of luminance and chrominance when decomposed into chrominance.

According to this, the color image data is divided into a plurality of pixel blocks, and the information of each component of the color image data is converted into a representative value and a coefficient representing a fine structure for each pixel block, and all the information is converted into a coefficient. Coding using a representative value of a component, a coefficient representing a fine structure of only a specific component, and a coefficient representing a relationship between a coefficient representing a fine structure of a component other than the specific component and a coefficient representing a fine structure of the specific component By doing so, the color image data is compressed. As described above, since the correlation of the color image data exists not only between the luminance and the color difference but also between the colors, these are collectively referred to as components, and by using the correlation of the components, the color conversion can be performed without trouble. Thus, color image data can be efficiently encoded.

A color image compression method according to a second aspect is the color image compression method according to the first aspect, wherein the contents of the component are luminance information and color difference information. According to this, by making the content of a component into luminance information and color difference information, instead of always using a high frequency coefficient of luminance (synonymous with an AC component) as a reference as in the related art, a high frequency coefficient of , The high-frequency coefficient of luminance can be extended.

According to a third aspect of the present invention, there is provided a color image compression method according to the first or second aspect, wherein the specific component is adaptively selected. The above-mentioned “selecting a specific component adaptively” means that the specific component is adaptively selected according to the value itself of the component, or the specific component is adaptively selected depending on the region such as an achromatic region or a chromatic region. Selecting is included. According to this, the color image data can be efficiently compressed by adaptively selecting the specific component as the reference coefficient.

According to a fourth aspect of the present invention, in the color image compression method according to any one of the first to third aspects, the coefficient representing the fine structure of each component is larger than that of the original image. The control is performed so as not to have a sharp density change. According to this, image quality degradation can be suppressed by controlling so that the coefficient representing the fine structure of each component does not have a sharper change in density than the original image.

According to a fifth aspect of the present invention, in the color image compression method according to any one of the first to fourth aspects, a low frequency coefficient subjected to sub-band conversion is used as the representative value. A sub-band-converted high-frequency coefficient is used as a coefficient representing the fine structure.

According to this, the sub-band-converted low-frequency coefficient is used as the representative value, and the sub-band-converted high-frequency coefficient is used as the coefficient representing the fine structure. It is possible to express the correlation of information and efficiently express the coefficient itself representing the fine structure.

According to a sixth aspect of the present invention, there is provided a color image compression method according to any one of the first to fourth aspects, wherein a sub-band-converted low frequency coefficient is used as the representative value. As a coefficient representing the fine structure, a vectorized high-frequency coefficient obtained by sub-band conversion is used.

According to this, low frequency coefficients obtained by sub-band conversion are used as representative values, and vectorized high-frequency coefficients obtained by sub-band conversion are used as coefficients representing a fine structure. It is possible to efficiently encode, and it is possible to efficiently represent the correlation of the fine structure information between the components and efficiently express the coefficient itself representing the fine structure.

According to a seventh aspect of the present invention, there is provided a color image compression method according to any one of the first to sixth aspects, wherein each of the coefficients is changed according to a sharp change in density in the pixel block. The fixed-length compression is performed by adaptively changing the number of bits allocated to. According to this, efficient fixed-length compression can be achieved by performing fixed-length compression by adaptively changing the number of bits allocated to each coefficient according to the intensity of the density change in the pixel block. .

According to an eighth aspect of the present invention, there is provided a color image compression method according to any one of the first to seventh aspects, wherein the coefficient representing the fine structure other than the specific component and the fineness of the specific component are different. The coefficient representing the relationship with the coefficient representing the structure is a coefficient representing the similarity between the coefficient representing the fine structure other than the specific component and the coefficient representing the fine structure of the specific component.

The use of the "coefficient representing the relationship between the coefficient representing the microstructure other than the specific component and the coefficient representing the microstructure of the specific component" means, for example, the use of the following. That is, the ratio between the two is used. Only when both microstructures are proportional, the ratio between them is used, otherwise, nothing is used, or information indicating that components other than the specific component are flat in the pixel block is used. As a first coefficient example, a ratio between the two is sent, and as a second coefficient example, a difference between a coefficient representing an actual microstructure and a coefficient representing a microstructure that can be extended from the ratio of both is used. As a first coefficient example, the ratio between the two is used only when the fine structures are proportional to each other, otherwise nothing is used, or a component other than the specific component is flat within the pixel block. Using information, the second
As a coefficient example, a difference between a coefficient representing an actual microstructure and a coefficient representing a microstructure that can be extended from the first coefficient example is used. A rotation angle and a magnification that bring the two closest to each other are used. The rotation angle, magnification, and difference between the two which are closest to each other are used. A first-order transformation matrix that is closest to both is used. The quantized coefficients listed above are used. There are things.

When the above-mentioned "coefficient representing the relationship between the coefficient representing the fine structure other than the specific component and the coefficient representing the fine structure of the specific component" is prepared, the coefficient is prepared for each component other than the specific component. And the case where one coefficient is created collectively for all components other than the specific component, and the case where the components other than the specific component are divided into several groups and the coefficient is created for each group Also included.

The above “coefficient representing similarity” includes the following. That is, when a coefficient representing the fine structure other than the specific component is Hr and a coefficient representing the fine structure of the specific component is Hg, | k
K such that Hr-Hg | is minimized, a value obtained by quantizing the above-described k, and | kHr-Hg when the coefficient representing the fine structure other than the specific component is proportional to the coefficient representing the fine structure of the specific component. Is a proportional constant k such that | is minimized. In other cases, a coefficient representing that the two are not in a proportional relationship, a coefficient obtained by quantizing the above k, and the like are included.

According to this, the coefficient representing the relationship between the coefficient representing the fine structure other than the specific component and the coefficient representing the fine structure of the specific component includes the coefficient representing the fine structure other than the specific component and the fine structure of the specific component. Since the coefficient representing the degree of similarity to the coefficient representing the color image data, the compression efficiency of the color image data can be improved.

According to a ninth aspect of the present invention, there is provided a color image compression apparatus for compressing color image data composed of a plurality of components, wherein the color image data is nxm (n, m Means for dividing into pixel blocks of a natural number), and means for converting the information of each component into a representative value and a coefficient representing a fine structure for each pixel block. Using a representative value of the component, a coefficient representing the fine structure of only the specific component, and a coefficient representing the relationship between the coefficient representing the fine structure of the component other than the specific component and the coefficient representing the fine structure of the specific component. It becomes something.

According to this, the color image data is divided into a plurality of pixel blocks, and the information of each component of the color image data is converted into a representative value and a coefficient representing a fine structure for each pixel block, and all the components are converted to a representative value. Coding using a representative value of a component, a coefficient representing a fine structure of only a specific component, and a coefficient representing a relationship between a coefficient representing a fine structure of a component other than the specific component and a coefficient representing a fine structure of the specific component By doing so, color image data can be efficiently encoded.

[0027]

Embodiments of the present invention will be described below in detail with reference to the drawings. Embodiment 1 FIG. In the first embodiment, taking a printer as an example, a case will be described in which color image data is reversibly compressed in order to reduce the transfer time when an image is transferred from a host computer to a printer. Here, the reversible compression means a compression in which image information can be completely restored before and after the compression, that is, a lossless compression.

In the first embodiment, the compression process is performed as follows without converting the color of an 8-bit color image of three colors RGB. Two-layer subband conversion is performed for each color. The G color image is encoded independently as a specific component. The low frequency coefficients of the R color and the B color are independently encoded. Regarding the high-frequency coefficients of R and B colors, first, a high-frequency coefficient of G color and a high-frequency coefficient vector of each of R and B colors are created, a ratio between the high-frequency coefficient vectors is calculated, and information representing the ratio is quantized. And encode. Next, a difference value between the high frequency coefficients of R and B colors to be decoded from the information indicating the quantized ratio and the actual high frequency coefficients of R and B colors is encoded.

Here, the G color is "specific component".
, The low-frequency coefficient of the sub-band conversion corresponds to the “representative value”, the high-frequency coefficient corresponds to the “coefficient representing the fine structure”, and the difference value described above is “ Coefficient representing the relationship between the coefficient representing the microstructure and the coefficient representing the microstructure of the specific component. "

Next, the above-mentioned processes (1) to (4) will be described individually. First, the subband transform is a broad concept including not only orthogonal transform such as DCT and Fourier transform but also overlap transform.

FIG. 1 is a diagram for explaining the Haar wavelet transform. In the first embodiment, the Haar wavelet transform shown in FIG. 1 is used as the sub-band transform. This Haar wavelet transform refers to pixels a, b, and
This is a conversion for converting the low frequency coefficient into four coefficients LL, HL, LH, and HH using c and d.
H and HH represent high frequency coefficients.

The meaning of the value enclosed by [] shown in FIG. 1 is replaced by [a] here.
It is assumed that the value of [a] represents an integer closest to a among numbers smaller than (a + 1). In such a Haar wavelet transform, the original pixel value can be completely restored by the inverse transform.

In the first embodiment, this subband conversion is performed over two layers as shown in FIG. For example, as shown in FIG. 3, the meaning of the two hierarchies is to take a pixel block every 4 × 4 block,
X4 is divided into four 2 × 2 sub-blocks and subjected to the Haar wavelet transform (first layer). The four LL coefficients obtained by this are collected and the Haar wavelet transform (second layer) is performed again. means.

To facilitate understanding, the coefficients obtained by the conversion in units of sub-blocks of 2 × 2 pixels are denoted by LL1 and HL.
1, LH1, HH1 coefficients (coefficients of the first layer), and the coefficients obtained by the conversion in the second step are LL2, HL2, L
They are called H2 and HH2 coefficients (coefficients of the second hierarchy) (see FIG. 2).

As a result, the information in the 4 × 4 pixel block includes one low frequency coefficient LL2 and HL as the high frequency coefficient.
2, LH2, HH2 coefficient is one each, HL1, LH1,
The HH1 coefficient is divided into four 16 coefficients each (see FIG. 3: HL1, LH1, and HH1 are indexed for each sub-block).

The above-mentioned coding means entropy coding. The entropy coding includes Huffman coding, Lang-length coding, and the like. The entropy coding is described in detail in various documents (for example, "Image Information Processing" by Aisin, written by Nakajima, and Morikita Publishing). Omitted.

In the above description, the high-frequency coefficient as a "coefficient representing a fine structure" is created, but the high-frequency coefficient of the subband conversion coefficient is very suitable for representing the local structure of an image. However, when considering the similarity relation of the entire microstructure, it is more efficient to calculate the ratio by vectorization than to calculate the ratio for each coefficient. The reason is,
For example, (HL1, L
This is because, when H1 and HH1) are created, the vector itself is often proportional to the high frequency coefficient vector of the G color. As an example, in a pixel block having an edge in the vertical direction, as is clear from the Haar wavelet transform equation in FIG. 1, only HL has a large value in R, G, and B colors, and LH, HH Has only a value close to zero. That is, the high-frequency vectors of the R, G, and B colors each have a substantially (x, 0, 0) shape, and are in a proportional relationship as a whole vector.

In the first embodiment, as shown in FIG. 4, a 15-dimensional vector composed of three high-frequency coefficients of the second hierarchy and twelve high-frequency coefficients of the first hierarchy is created as a high-frequency coefficient vector, First, it is checked whether or not the coefficient vector is proportional to the high frequency coefficient vector of G color. Regarding whether or not the value is proportional, the following expression (1), which is the cosine of the angle between the two 15-dimensional vectors, is calculated. When the absolute value of this value is close to 1, it is determined that the value is proportional.

[0039]

(Equation 1)

In the above equation (1), Hr is a 15-dimensional high frequency coefficient vector of R color or B color, and Hg is G
These are the 15-dimensional high-frequency coefficient vectors of the colors, whose denominator is their absolute value. The function of equation (1) must be -1
Takes a value between 1 and 1. If 1 or -1, the two vectors are proportional.

The determination as to whether or not the absolute value is close to 1 is made by setting an appropriate threshold value and comparing it. If the result of this determination is that the two vectors have a proportional relationship (Hr = kHg) or a pseudo-proportional relationship (here, denoted by Hr ≒ kHg), the proportional value k is quantized and encoded. .

In the first embodiment, k is -2.0,-
It is quantized to 7-value 3-bit values of 1.0, -0.5, 0, 0.5, 1.0, and 2.0. By this quantization, for example, k =
3 is quantized with the closest k = 2. In addition, as a result of the above determination, k = 0 for a pixel block for which it has been determined that the high-frequency vector is not proportional.

Then, the difference value (Hr-kHg) between the actual high frequency coefficient vector Hr and kHg is encoded. When k = 0, the difference value is, of course, the value of the high frequency coefficient of the R or B color itself. The above description is the above processing, and these processings are further performed for each of the R color and the B color.

Next, a description will be given of a color image compression apparatus for compressing the above-described color image data. FIG. 5 is a block diagram of a color image compressing apparatus that performs a compression process according to the first embodiment. FIG. 6 is a block diagram of a high-frequency coefficient encoding unit other than the specific components in FIG. In FIGS. 5 and 6, it is assumed that the lines where two or more lines intersect are connected only at the portions with black circles, and the “specific C” in FIG. 5 is the “specific component”. Means.

4 × 4 buffers 501 and 5 for each color shown in FIG.
At 11,521, the host computer reads color image data of 4 × 4 pixel blocks for each color into a buffer. Then, the color image data of each color is converted into a low-frequency coefficient of each color and a high-frequency coefficient of G color as a specific component by the sub-band conversion units 502, 512, and 522 in the entropy coding unit 50 in the next stage.
4 to be coded.

The high frequency coefficients of the R and G colors are converted into the proportional value (k) and the difference value (Hr-kHg) by using the high frequency coefficients of the G color in the high frequency coefficient encoding units 513 and 523 other than the specific component. ) Is sent to the entropy encoding unit 504 and encoded. High frequency coefficient encoding units 513 and 52 other than specific components
3, the input G color, as shown in FIG.
The high frequency coefficient of R color (or B color) is calculated by the proportional value calculation unit 60.
In step 1, a proportional value k is calculated using the above equation (1), and the difference value calculating unit 60 is calculated using the proportional value k and the high frequency coefficient.
In 2, the difference value (Hr-kHg) is calculated.

Thereafter, the color image data encoded by the entropy encoding unit 504 is sent to a printer (not shown) and decompressed by the printer.

Next, effects of the first embodiment will be described. The RGB three-color image data of the first embodiment is used for restoring the low-frequency coefficients and the high-frequency coefficients of G and the high-frequency coefficients of R and B for each of the 4 × 4 pixel blocks. Are respectively decomposed into a proportional value k and a difference value. Therefore, the code amount for each low-frequency coefficient and the high-frequency coefficient of G color is not different from the code amount in the case of compressing a normal sub-band transform coefficient, but when the high-frequency coefficients of R and B colors are directly encoded. Then, it is verified whether there is a difference in the code amount between the case where the coding is performed by dividing into the proportional value and the difference value.

First, as for the proportional value k, since the amount of information is at most about 3 bits for every 16 pixels, it can be said that the amount of information is very small.

Next, regarding the difference value (Hr-kHg),
In the region where Hr is proportional to Hg, it is clear that (Hr−kHg) has a smaller range of numerical values than the original Hr, and the data in which the numerical values are localized is obtained by entropy coding. Can be highly compressed. Also, for a region where Hr is not proportional to Hg, k
= 0, and Hr sends the value of Hr as it is, so the code amount increases by the amount that the value of k (= 0) must be sent. However, the area that is not in a proportional relationship is only an area in which three or more colors having greatly different density values are mixed in a 4 × 4 pixel block.

This example will be described with reference to FIG. FIG.
(A)-(c) is a figure which shows the example of a coefficient conversion regarding the effectiveness of this Embodiment 1. The original image 701 and the conversion coefficient 702 in FIG. 7A correspond to the relationship between the original image and the finally obtained conversion coefficient in FIG.

The first image example 703 and the first coefficient example 704 shown in FIG. As shown in FIG. 13B, in a region where only two colors or less exist in the pixel block, all high-frequency coefficients except for LL2 are:
K = 0.5 for the R color and k = 0 for the B color, which is proportional to Hg.

FIG. 7 shows an example in which three colors whose density values are mutually increased in a pixel block.
A second image example 705 and a second coefficient example 706 of (c) are shown. In this case, the high frequency coefficients are not proportional to each other as shown in the second coefficient example 706. As described above, a region where the high-frequency coefficient vector is not in a proportional relationship is a region where three or more colors having greatly different density values are mixed in the pixel block. In general, there are very few cases where three or more colors having greatly different density values are mixed in a small area such as a 4 × 4 pixel block. From this, it can be said that the high-frequency coefficient vector is in a proportional relationship in most regions in the image, and thus the code amount at the time of compression as a whole is reduced by the present invention.

The example of the original image 701 in FIG.
The above description is an example of a case where the inside of a pixel block is completely two colors. However, for example, data such as scan data that fluctuates in the value of an image (for example, data “R = 50” is “54”)
Or "48").
If the fluctuation is such that the human eye can recognize that only two colors exist in the pixel block, it can be said that the high-frequency coefficient vector is approximately in a proportional relationship. That is, since the absolute value of the value of the above equation (1) is sufficiently close to 1, the code amount can be reduced as well.

As described above, according to the first embodiment, since the compression efficiency of the color image data can be improved, the data transfer time and the like can be shortened. Further, according to the first embodiment, since the high-frequency coefficient vector has a proportional relationship as a whole vector, and the code amount at the time of compression can be reduced, the proportional relationship can be efficiently expressed.

In the first embodiment, the G color is always used as the specific component as a method of selecting the specific component. However, even if this color is replaced with the R color or the B color, the same advantageous effects can be obtained. it can.

Further, in the case where the specific component is not fixed but is adaptively selected and changed, for example, it is possible to perform switching in units of pixel blocks. As an example in which switching is performed in units of pixel blocks, there is a case where, of the three colors of RGB, the one with the most intense density change in a pixel block is set as a specific component and compressed together with a flag indicating the specific component. In this case, the value of the proportional value k is always 1 or less. As described above, in the first embodiment, these cases are included as a method of selecting a specific component.

Further, in the first embodiment, RG
The correlation between the respective colors of B is directly used, but this is performed using the data after the color conversion. For example, in the YCrCb color conversion, the high frequency coefficient vector of Cr and Cb is proportional to the high frequency coefficient vector of Y. A similar theory holds. In this case, as to how to select a specific component, the theory described in the second embodiment can be applied as it is.

Embodiment 2 In the second embodiment, taking a digital copying machine as an example, a case will be described in which information for one image is fixed-length compressed and stored in a memory (hereinafter referred to as a page memory) for high-speed copying. Here, the fixed-length compression refers to a compression method in which the length of a generated compression code is the same for each pixel block, and means a compression method that facilitates memory access. This fixed-length compression is compression in which image information deteriorates before and after compression, that is, irreversible compression.

In the second embodiment, the low-frequency coefficient LL is used as the “representative value”, the high-frequency coefficient vectors HL, LH, HH are used as the “coefficient representing the fine structure”, and the “fine structure of components other than the specific component” is used. The proportional value k is used as the "coefficient representing the relationship between the coefficient representing the fine structure of the specific component" and the coefficient representing the fine structure of the specific component.

In the second embodiment, since a digital copying machine processes an 8-bit color image of three colors of RGB in one scan, the color image of three colors of RGB is temporarily compressed and stored in a page memory after the scan, and then is compressed. Create CMYK versions sequentially while decompressing.

FIG. 8 is a block diagram showing a schematic configuration of a digital copying machine used in the second embodiment. As shown in FIG. 8, the RGB input to the compression unit 801 in one scan
Are fixed-length compressed by the compression unit 801 and stored in the page memory 802. If the user wants to edit the image, the color image data is sent from the page memory 802 to the editing panel 807, and the data reflecting the edited content is stored again in the page memory 802. As described above, fixed-length compression is advantageous for data compression in editing.

At the time of writing the color image data, the color image data read from the page memory 802 is sent to the decompression unit 803 and decompressed by RGB, and the decompressed RGB data is converted to the CMYK conversion unit 80.
In step 4, the data is first converted into a C-plate, filtered by a filter unit 805, and sent to a writing unit 806. After the C plane is written, the color image data is read out again from the page memory 802, and after the same processing as described above,
A version is written. Furthermore, similarly, Y version and K version
The processing of the plate follows.

The digital copying machine according to the second embodiment is configured as described above, and the compression processing of the color image data performed by the compression section 801 therein will be described in detail below. A specific procedure for compressing color image data is to read an image for each 2 × 2 pixel block, perform color conversion on the image, and then select one specific component from luminance and color difference. Both specific colors and non-specific colors are subjected to one-layer subband conversion for each pixel block. Using the coefficients obtained by the sub-band conversion, the image area is divided into an area where the density changes drastically (hereinafter referred to as an edge area) and another area (hereinafter referred to as a non-edge area). The low frequency coefficient is adaptively quantized for each region. For the high frequency coefficient of the specific component, a three-dimensional high frequency coefficient vector (HL, LH, HH) is created. For the high-frequency coefficient of the non-specific component, the proportional value is quantized and encoded for each region using a proportional relationship with the high-frequency coefficient vector of the specific component. Coefficients representing different regions are also encoded, and fixed-length compression is performed on a pixel block basis as a whole of luminance and chrominance. When performing the above processing, in order to suppress image quality deterioration, the proportional value is suppressed so that the high-frequency coefficient quantized at the time of encoding the high-frequency coefficient does not show a steep change in density than the original value.

FIG. 9 is a block diagram showing the configuration of the compression section 801 in FIG. Note that the intersecting lines in FIG. 9 are connected only by black circles. The compression unit shown in FIG.
Each color (R, G, B) 2 × 2 buffers 901 and 91
1,921, a color conversion unit 902, a specific C (component) sub-band conversion unit 903, a non-specific C (component) sub-band conversion unit 913, 923, an area determination unit 90
4. High frequency coefficient quantization section 905, proportional value calculation section 915,
925, and adaptive quantization / fixed-length code generation unit 906
Etc.

Here, the correspondence relationship in the case where the above processes (1) to (4) are performed by each unit in FIG. 9 will be described below. In each color 2 × 2 buffer 901, 911, 921 in FIG.
The color image data read into the buffer for each pixel block is color-converted (processed) by the color conversion unit 902,
At the same time, one of the luminance and the color difference is selected as a specific component (how to select the specific component will be described later in detail). Here, regarding color conversion, YCr
Many color conversions such as Cb, Lab, and YIQ are known, but in the second embodiment, the color conversion according to the following equation (2) will be used.

[0067]

(Equation 2)

After performing the color conversion as described above, the specific component is subjected to sub-band conversion in the specific C sub-band conversion unit 903, and the non-specific component is converted to the non-specific C
Subband conversion is performed in the subband conversion units 913 and 923 (processing). The Haar wavelet transform described above (see FIG. 1) is used for this sub-band transform. In this case, the transform coefficients obtained because of the one-layer sub-band transform are only the four coefficients shown in FIG.

Next, the area determination unit 904 determines whether the pixel block of interest is an edge area or a non-edge area using the high frequency coefficient of the specific component (processing). In this case, the absolute value of the high-frequency coefficient vector (HL, LH, HH) 用 い (HL ² + LH ² + H) is determined by using the conversion coefficient of the specific component.
If H ² ) is 16 or more, the region is defined as an edge region, but is not necessarily limited to this. The result of the area judgment is
The result is sent to the adaptive quantization / fixed-length code creation unit 906.

The low frequency coefficient of each component is
The high-frequency coefficient of the specific component is sent to the adaptive quantization / fixed-length code creation unit 906 as it is, and is quantized as a high-frequency coefficient vector in the high-frequency coefficient quantization unit 905 (processing). Specific high-frequency coefficient vectors (HL,
In the quantization of (LH, HH), the high-frequency coefficient vector of the specific component is quantized into 31-valued 5-bit data as shown in FIG.

Here, when image data is compressed into a fixed number of bits as in fixed-length compression, creating a high-frequency coefficient vector depends on the proportion of the entire vector as described in the first embodiment. In addition to being able to use relationships, it is significant in that it can be coded efficiently. That is, generally, the edges of an image are locally vertical edges or horizontal edges, and the two edges are rarely mixed. This is particularly true for 2 × 2 pixel blocks. If this tendency is used, for example, a high-frequency coefficient vector (for example, 100, 100, 0, etc.) in which HL representing a vertical edge and LH representing a horizontal edge both have large values is excluded, and the existence probability is large (0, x, 0), (x,
High frequency coefficient vectors of the form (0,0), (0,0, x) can be effectively sampled. The example of the high-frequency coefficient vector shown in FIG. 10 is an example of such a high-frequency coefficient vector.

As for the high frequency coefficient of the non-specific component, a proportional value is calculated by comparing the high frequency coefficient of the specific component and the high frequency coefficient of the non-specific component in the proportional value calculation units 915 and 925, and the proportional value is quantized. Encode (process). The method of calculating the proportional value is performed by using the above equation (1) in the same manner as in the first embodiment, and the proportional value k is −2.0, -1.0, -0.5, 0,
It is quantized to 7-value 3 bits of 0.5, 1.0, and 2.0.
Note that, for a non-proportional region, only k = 0 is sent and no difference is sent. The proportional value k thus obtained is sent to the adaptive quantization / fixed-length code creation unit 906.

Finally, the adaptive quantization / fixed-length code generation unit 906 receives the area determination result, the low-frequency coefficient, the quantized high-frequency coefficient vector of the specific component, and the proportional value, and receives the area determination result. Quantization and fixed-length encoding are performed with reference to these. Specifically, one bit of flag bits representing an area (edge / non-edge) is created, each transform coefficient is quantized, and a combination of the two is used to create a fixed-length code as follows (process).

That is, in the edge area, the low frequency coefficient LL is set to 4 for both the specific component and the non-specific component.
The bits are quantized (processed), and the high-frequency coefficient of the specific component encodes the 5-bit 31 value obtained by the process, and the proportional value of the non-specific component encodes the 3-bit 7 value of each component obtained by the process as it is. as a result,
The code length of the edge area is 2 in total including the flag bit.
It becomes 4 bits.

In the non-edge area, the low-frequency coefficient LL of the specific component is not quantized, but remains at 8 bits, and the LL of the non-specific component is quantized to 7 bits (processing). For the high frequency coefficient vector, (HL, LH, HH) = (0, 0, 0) for each component
It does not send any code. As a result, the code length of the non-edge area is 23 bits in total including the flag bits.

As described above, the image information of the 2 × 2 pixel block is compressed to a fixed length of 24 bits (compressed to a quarter). The compressed fixed-length code is then stored in the page memory 802, and is used in the above-described processing at the next stage and thereafter.

FIG. 11 is a diagram showing a breakdown of fixed-length codes according to the second embodiment. At the time of decoding, first, area determination is performed based on the value of the flag bit, and then, each subband coefficient is obtained in accordance with the arrangement of codes in FIG. The difference between the quantization methods in the respective regions is that the resolution is emphasized in the edge region and the gradation is emphasized in the non-edge region.

Further, the processing is performed as follows. First, when the processing (processing) of the high frequency coefficient quantization unit 905 in FIG. 9 is performed, a vector search is performed so that the high frequency of the compressed image does not have a value greater than the high frequency of the original image. For example, a vector search is performed within a range in which the absolute value √ (HL ² + LH ² + HH ² ) of the quantized high frequency coefficient vector does not become larger than the absolute value of the high frequency coefficient vector of the original image.

Next, when the processing (processing) of the proportional value calculating sections 915 and 925 is performed, for example, when the proportional value before quantization is 0.9, the proportional value is quantized to a proportional value 1.0. Then, since the high-frequency coefficient after compression has a larger value than the high-frequency coefficient before compression, the density change in the pixel block becomes severe. Therefore, always quantize to a small value,
Here, quantization is performed to 0.5. The reason why such processing (processing) is performed is that when the density change becomes more intense than that of the original image, the compressed image may be rough or blurry as compared with the case of no compression, and is visually noticeable. This is because image quality deteriorates. On the other hand, in the case of deterioration in which the change in density is smaller, the deterioration is generally not visually noticeable. In the second embodiment, unlike the first embodiment, the difference value is not encoded.
This processing is required.

However, even in the case of the first embodiment, if the memory size of the printer is fixed, if the compressed image cannot fit in the memory, the difference value is truncated and the lossy compression is performed. This processing may be performed, and in this case, this processing may be necessary to improve the image quality of the compressed image.

Next, how to select a specific component in the second embodiment will be described. Since the high frequency coefficient of the specific component is quantized more accurately using the high frequency coefficient vector instead of the proportional value, it is desirable to select information having the most important high frequency coefficient as the specific component. For example, a component having a visually large influence or a component having the largest edge. Based on this, the method of selecting the specific component includes all the following cases in the second embodiment. (1) When always fixing a specific component Brightness is always used as a specific component. (2) When the specific component is adaptively switched Among the luminance / color difference (YUV) in the target pixel block,
The component having the largest absolute value of the low frequency coefficient is defined as the specific component. Among the luminance / color differences (YUV), those having the absolute value with the largest gradient in the pixel block are defined as specific components. Only when the high frequency coefficient of the color difference and the high frequency coefficient of the luminance are proportional and the color difference has a gradient greater than the luminance, the color difference is set as the specific component, and in other cases, the luminance is set as the specific component. The edge area is divided into a color edge area and other edge areas, and the color difference is a specific component for the color edge area, and the luminance is a specific component for the other edge areas.

Here, in the case of using the item (2), since the decoding side determines which coefficient of YUV is a specific component, a code length of 1 to 2 bits is required. On the other hand, in the case of the item (2), a specific component can be adaptively selected without an increase in code length.

As an example where the item (2) is effective, there is, for example, a case as shown in FIG. 12 showing an example of how to select a specific component. As shown in FIG.
When only the color has a large value in the pixel block, U has a larger edge than Y, and the absolute value of the low frequency coefficient is also larger in U. In such a case, it is possible to store edges more accurately if U is a specific component than to Y, and it is not necessary to densely prepare one or more proportional values k.

On the other hand, as a specific embodiment of the item (2), for example, the magnitude of the gradient is determined by referring to the absolute values of the high frequency coefficient vectors of the color difference and the luminance. Further, as a specific example of the item (2), the determination as to whether or not a color edge occurs is performed by using a region prepared for adaptive filtering, adaptive gradation processing, and the like on the digital copying machine side. The judgment mechanism is used as it is.

In the second embodiment, the data after color conversion is described. However, the same theory can be applied to the case where RGB colors are directly subjected to fixed-length coding. How to select a specific component in this case is
The theory on how to select a specific component described in the first embodiment can be applied as it is.

Next, effects of the second embodiment will be described. The advantage of making the coefficient representing the fine structure other than the specific component proportional to the coefficient representing the fine structure of the specific component as described in the second embodiment has been described in the first embodiment. In the case of 2, since a very small pixel block of 2 × 2 is used, it is extremely rare that three or more colors that are greatly different from each other exist in the pixel block, and the compression efficiency can be improved.

According to the second embodiment, it is possible to efficiently compress data by utilizing the fine structure correlation between components or between luminance and chrominance, thereby reducing the memory capacity. Transfer time and the like.

According to the second embodiment, the most important fine structure information (in this case, edge information) can be stored more accurately by selecting a specific component adaptively. Since adaptive component selection is performed by entropy coding, it is more effective in fixed-length compression as in the second embodiment than in variable-length compression in which it is necessary to increase the correlation with surrounding pixel blocks. Can be demonstrated.

Furthermore, according to the second embodiment, at the time of vector search and the calculation of the proportional value, control is performed so that the coefficient representing the fine structure of each component does not have a sharper change in density than the original image. It is possible to eliminate the roughness and blur of the image and prevent the image quality from being visually noticeable.

According to the second embodiment, high-frequency coefficients of sub-band conversion are vector-quantized,
It is possible to efficiently represent the correlation of the fine structure information between the components, and it is also possible to efficiently represent the coefficient itself representing the fine structure.

Further, according to the second embodiment, efficient fixed-length compression can be achieved by adaptively performing fixed-length encoding of these subband coefficients and proportional values for each region.

In the second embodiment, the color conversion is performed using the above equation (2). However, the present invention is not limited to this, and general color conversion such as YCrCb, Lab, YIQ, etc. is used. It is also possible to obtain a suitable effect in such a case.

[0093]

As described above, according to the color image compression method of the present invention (claim 1), the correlation of color image data exists not only between luminance and color difference but also between each color. Since the correlation of the components is used, it is possible to efficiently encode and compress the color image data without taking the trouble of color conversion.

Further, according to the color image compression method of the present invention (claim 2), since the content of the component is the luminance information and the color difference information, the high frequency coefficient of the luminance is set as necessary with reference to the high frequency coefficient of the color difference. Can be extended.

Further, according to the color image compression method of the present invention (claim 3), a specific component as a reference coefficient is adaptively selected, so that color image data can be efficiently compressed. Can be.

According to the color image compression method of the present invention (claim 4), the coefficient representing the fine structure of each component is controlled so as not to have a steep change in density as compared with the original image. Image quality deterioration can be suppressed.

According to the color image compression method of the present invention (claim 5), a sub-band-converted low-frequency coefficient is used as a representative value, and a sub-band-converted high-frequency coefficient is used as a coefficient representing a fine structure. As a result, the correlation of the fine structure information between the components can be efficiently represented, and the coefficient itself representing the fine structure can be efficiently represented.

According to the color image compression method of the present invention (claim 6), a low-frequency coefficient obtained by sub-band conversion is used as a representative value, and a high-frequency coefficient obtained by sub-band conversion is used as a coefficient representing a fine structure. Since the vectorized version is used, more efficient encoding is possible, and the correlation of the fine structure information between each component can be efficiently represented, and the coefficient itself representing the fine structure can be efficiently used. Can express well.

Further, according to the color image compression method of the present invention (claim 7), the fixed length compression is performed by adaptively changing the number of bits allocated to each coefficient according to the intensity of the density change in the pixel block. Since this is performed, efficient fixed-length compression can be achieved.

Further, according to the color image compression method of the present invention (claim 8), the coefficient representing the relationship between the coefficient representing the fine structure other than the specific component and the coefficient representing the fine structure of the specific component is changed to the coefficient other than the specific component. Since the coefficient representing the similarity between the coefficient representing the fine structure of the specific component and the coefficient representing the fine structure of the specific component, the compression efficiency of the color image data can be improved.

Further, according to the color image compression apparatus of the present invention (claim 9), the representative values of all the components,
A coefficient representing the microstructure of only certain components,
Since the encoding is performed using the coefficient representing the fine structure of the component other than the specific component and the coefficient representing the relationship between the coefficient representing the fine structure of the specific component, the color image data is efficiently encoded and compressed. Can be.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a Haar wavelet transform according to the first embodiment.

FIG. 2 is a diagram illustrating subband conversion performed hierarchically in the first embodiment.

FIG. 3 is an explanatory diagram of coefficient transformation by Haar wavelet transformation performed over two layers in the first embodiment.

FIG. 4 is a diagram showing a 15-dimensional vector including three high-frequency coefficients in the second hierarchy and 12 high-frequency coefficients in the first hierarchy as high-frequency coefficient vectors in the first embodiment.

FIG. 5 is a block diagram illustrating a color image compression apparatus that performs a compression process according to the first embodiment;

FIG. 6 is a block diagram of a high-frequency coefficient encoding unit other than the specific components in FIG. 5;

FIG. 7 is a diagram showing an example of coefficient conversion relating to the effectiveness of the first embodiment.

FIG. 8 is a block diagram showing a schematic configuration of a digital copying machine used in the second embodiment.

FIG. 9 is a block diagram illustrating a configuration of a compression unit in FIG. 8;

FIG. 10 is a diagram in which a high-frequency coefficient vector of a specific component is quantized into 31-valued 5 bits.

FIG. 11 is a diagram showing a breakdown of fixed-length codes according to the second embodiment.

FIG. 12 is a diagram illustrating an example of how to select a specific component.

[Description of Code] 501 G color 4 × 4 buffer 502, 512, 522 Subband conversion unit 504 Entropy coding unit 511 R color 4 × 4 buffer 521 B color 4 × 4 buffer 513, 523 High frequency coefficient coding other than specific components Unit 601 proportional value calculation unit 602 difference value calculation unit 701 original image 702 conversion coefficient 703 first image example 704 first coefficient example 705 second image example 706 second coefficient example 801 compression unit 802 page memory 803 decompression unit 804 CMYK conversion unit 805 Filter unit 806 Writing unit 807 Editing panel 901 G color 2 × 2 buffer 902 Color conversion unit 903 Specific component subband conversion unit 904 Area judgment unit 905 High frequency coefficient quantization unit 906 Adaptive quantization / fixed length code generation unit 911 R Color 2 × 2 buffer 913, 923 Non-specific component subband converter 915, 925 Proportional value calculator 921 B color 2 × 2 buffer

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04N 7/30 H04N 7/133 Z 5J064 9A001 (72) Inventor Yukiko Yamazaki 1-3-3 Nakamagome, Ota-ku, Tokyo No. 6 F-term in Ricoh Co., Ltd. (reference) PP31 PP32.

Claims (9)

[Claims] 1. A color image compression method for compressing color image data composed of a plurality of components, the method comprising: dividing the color image data into n × m (n and m are natural numbers) pixel blocks. Converting the information of each component into a representative value and a coefficient representing a fine structure for each pixel block, when compressing the color image data, a representative value of all components, a specific value A color image encoded using a coefficient representing a fine structure of only a component, and a coefficient representing a relationship between a coefficient representing a fine structure of a component other than a specific component and a coefficient representing a fine structure of a specific component. Compression method. 2. The color image compression method according to claim 1, wherein the content of the component is luminance information and color difference information. 3. The color image compression method according to claim 1, wherein the specific component is adaptively selected. 4. The color according to claim 1, wherein the coefficient representing the fine structure of each component is controlled so as not to have a sharp change in density compared to the original image. Image compression method. 5. The sub-band-converted low-frequency coefficient is used as the representative value, and the sub-band-converted high-frequency coefficient is used as a coefficient representing the fine structure. 7. A color image compression method according to any one of the above. 6. The method according to claim 1, wherein a low-frequency coefficient obtained by sub-band conversion is used as the representative value, and a high-frequency coefficient obtained by sub-band conversion is vectorized as the coefficient representing the fine structure. A color image compression method according to any one of claims 1 to 4. 7. The fixed-length compression according to claim 1, wherein the number of bits allocated to each coefficient is adaptively changed according to the intensity of the density change in the pixel block. A color image compression method according to one of the above. 8. The coefficient representing the relationship between the coefficient representing the fine structure other than the specific component and the coefficient representing the fine structure of the specific component represents the coefficient representing the fine structure other than the specific component and the fine structure of the specific component. 8. The color image compression method according to claim 1, wherein the coefficient is a coefficient representing a degree of similarity with the coefficient. 9. A color image compression apparatus for compressing color image data composed of a plurality of components, comprising: means for dividing the color image data into n × m (n and m are natural numbers) pixel blocks. Means for converting the information of each component into a representative value and a coefficient representing a fine structure for each pixel block; and, when compressing the color image data, a representative value of all components and only a specific component. A color image compression apparatus encoding using a coefficient representing a fine structure of a component and a coefficient representing a relationship between a coefficient representing a fine structure of a component other than the specific component and a coefficient representing a fine structure of the specific component. .