JP3481644B2 - Image compression device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image compression apparatus for efficiently encoding an image having two or more layers of bit planes, for example, a multi-color image. 2. Description of the Related Art At present, an image (document image is simply referred to as a document image) or a computer graphic (CG is referred to as a CG in the text) made of a computer, such as a document, a diagram, and a table, is electronically recorded / stored / database. A system for searching for an arbitrary image at an arbitrary time has appeared. The image can be stored in a bitmap format, a character code, vector data, a font, control data, or the like. The former has advantages such that high-speed drawing can be performed because calculation for drawing is eliminated, and data can be shared. When saving images with bitmaps, compression technology that saves more images in less memory is essential, and for practical use,
It is necessary to use an inexpensive device and a short time required for search and reproduction. In order to extend a monochrome image from a binary image to a multivalued image, for example, a multicolor image of eight colors, it is necessary to extend one pixel to three bits. For example, R, G, and B are assigned one bit at a time. As a method of encoding the multi-color image, a method of applying an existing encoding technique to each of the R, G, and B planes has been proposed. When compressing an image color-expanded to eight colors, encoding is performed for each plane, so the number of pixels is three times as large as that of a binary image. Therefore, the information amount of the multi-color image after compression is significantly larger than that of the binary image. [0004] In "Color facsimile compatible with digital network using bubble jet recording and COSMIC coding", Vol. . The encoding means 3 in FIG. 8 is a pixel prediction means 11 using a Markov model.
And the dynamic arithmetic coding means 12. To explain the Markov model, which is an important concept of this method,
FIG. 9 is used. In FIG. 9, one certain bit plane,
For example, attention is paid only to the R plane. In the figure, a pixel to be coded is denoted by x, and its surrounding pixels a, b, c,
d, e, and f are referred to as reference pixels. The part (10) composed of the reference pixels is called a Markov model template. When pixel x is encoded in order 9 shown in the figure,
Each reference pixel has already been coded and can be referenced. The state determined by the values of the six reference pixels is referred to as a Markov state (64 states) in the text. The method using the Markov model utilizes the correlation between a coded pixel x and a reference pixel. Pixel prediction means 11
Predicts the coded pixel x from the Markov state. For example, when all the reference pixels are 0, there is a high probability that the coded pixel is also 0. Thus, prediction is possible. The value of the coded pixel is predicted from the Markov state, and when the predicted value matches the value of the coded pixel, the pixel is referred to as a superior pixel in the text, and is referred to as a recessive pixel if different from the prediction. In one image, the probability that the coded pixel is recessive (= 1−predominant probability) differs depending on the Markov state, but its value is small. As described above, the coding unit 3 predicts a coded pixel from a state of a reference pixel having a strong correlation with the coded pixel (a Markov state), and determines whether the coded pixel is inferior or superior. Arithmetic coding is performed using this judgment and the probability of occurrence of recessiveness determined by the Markov state. In FIG. 8, encoding is performed by the dynamic arithmetic encoding means 12. In dynamic arithmetic coding, the probability of this recessive occurrence is
It changes dynamically according to the frequency of occurrence of past recessiveness. The arithmetic code and the more detailed contents of this method are described, for example, in the chapter of binary coding in "International Standard for Multimedia Coding" (Maruzen) edited by Hiroshi Yasuda. [0007] In the above-mentioned system in which the Markov model and the dynamic arithmetic coding are combined, the color correlation between planes is used,
Further, the coding efficiency is improved. That is, not only pixels (12 pixels) in the plane but also encoded pixels in other planes are referred to as reference pixels. [0008] The prior art multi-color image coding method is intended for the target device and image. ・ For high-priced, high-quality devices such as color copiers and color facsimile machines. It is. The output means is paper. The image to be encoded is an already printed image or photograph. On the other hand, the target image and device in the present invention are as follows: The target image is a multi-valued image such as a multi-color image created by a computer or the like. The compressed image is stored in a recording device and output to a display or the like. The target devices are personal computers and personal portable devices. In view of the above, the coding method of the prior art is not optimized for the target device and image as the object of the present invention. In addition, the number of reference pixels is large, and the model template of the Markov model is different depending on the plane. The present invention provides the target image,
It is an object of the present invention to provide an image compression apparatus which can obtain a high compression ratio by an image encoding method suitable for the apparatus, and which can be realized at a low cost with a simple configuration and which has a short reproduction time. [0015] In order to solve the above-mentioned problems and to achieve the object, the present invention provides a method for printing characters, line drawings, and solid images.
Bits composed of image display elements and having two or more layers
This is a device for compressing multi-valued images with
An image compression apparatus comprising: a symbol converting means for converting the display symbols indicating the image display elements for the shown in coded symbol, and a compression means for coding the coded symbols for each bit plane, the The symbol conversion means is an n-bit display symbol indicating the image display element .
A conversion table for determining the Luo n-bit coded symbols
In the conversion table, each image display element
Properties of Binary Images Formed by Image Processing as 2D Images
Is the entropy determined from the correlation with the adjacent image
Calculation for all image display elements,
Rank in the order of larger entropy, and the above n images
For display elements with high entropy,
An image table having the smallest entropy is assigned to a coded symbol in which each bit is 1 and the other bits are 0.
Indicated element, n bits all 1
For an image showing the allocation and background, all n bits are 0
Are assigned to the coded symbols . (Embodiment 1) One embodiment of the present invention will be described using a map image (CG) as an example. Normally, bitmap maps include, for example, general roads, toll roads, rivers, place names,
A color is determined by a green area, a mark, and the like, and a symbol is determined for the color. In the text, the magnitude of the amount of information of a certain symbol (color) as an image is defined as the amount of information of an image made only of the symbol. In the image thus created, the statistical information of the image is completely different depending on the symbol, so that the image information amount of the symbol is also greatly different. For example, the amount of information as an image of a general road is very large, and the amount of information on green spaces, marks, and the like is extremely small. The information amount of an image obtained by combining two different symbols takes a value close to the sum (sum) of the information amounts of the images of the respective symbols. For example, the information amount of an image obtained by combining a road and a mark is close to the sum of the respective information amounts. The present invention focuses on these two points, and converts a color symbol (described later) into a coded symbol (described later) that provides the highest compression efficiency. Further, the present invention provides a method for configuring the symbol conversion means. The principle of the present invention will be specifically described with reference to FIG. The image information source 2 is a map image expanded by 3 bits in color. This can be expressed, for example, as three planes of R, G, and B shown in FIG. 5, and corresponds to R, G, and B in FIG.
In the text, (R, G, B) is a color symbol.
The color symbols are input to the symbol conversion means 1 which is the main element of the present invention, and are converted into symbols (D0, D1, D2) for encoding as shown in FIG. In the text, these are coded symbols. The method of configuring this symbol conversion means will be described later, but this greatly improves the compression ratio. The encoded symbol is input to the encoding means 3.
The encoding means 3 calculates each bit D of the encoded symbol.
0, D1, and D2. Bit plane (Fig. 5)
The data is encoded every time, and the compressed data 7 is generated. The encoding point order is, for example, as shown in FIGS.
Any order is acceptable. Here, (1), (2),
(3). . . Indicates the order of encoding. The encoding method is M
H, MR, MMR, arithmetic code, etc. may be used.
Also, the encoding method or parameters of each plane may be different. In the present embodiment, an encoding method in which a conventional Markov model and arithmetic encoding are combined is used. However, the use of the symbol conversion means greatly improves the compression ratio. In addition, since the compression ratio is high even if the number of reference pixels is reduced (even if color correlation between planes is not obtained), the size of the apparatus can be reduced. The compressed data 7 is recorded on the recording device 4. 4 may be a communication device. Next, the configuration of the symbol conversion means will be described. The construction method is to assign coded symbols to color symbols. As a pre-process, the information amount of an image formed by only one color symbol is obtained for all the color symbols (eight in this embodiment). The amount of image information of a color symbol is assumed to be an encoding method, and an amount indicating a statistical bias of an image, here, entropy is determined based on the encoding method, and this is set as the information amount. The method of calculating entropy will be described at the end of the first embodiment. After the completion of the previous stage, symbols are allocated according to the following procedure. Procedure 1: Each symbol is ranked in descending order of entropy. In this embodiment, eight color symbols are ranked. Procedure 2: 000 is usually assigned to the background having the highest entropy. Step 3: Except for the background, 001, 010, 1 are assigned to the top three symbols with high entropy.
Assign 00. The order is not specified in this embodiment. Step 4: 011 and 10 are assigned to the top three symbols with high entropy that have not yet been assigned.
1, 110 are assigned. The order is not specified in this embodiment. Step 5: Assign 111 to the symbol having the lowest entropy. In the above procedure, all 1s and 0s may be inverted. According to the construction method of the symbol conversion means proposed in the present invention, a color symbol having a large entropy is independently allocated to each bit plane of a coded symbol, and a small color symbol is allocated a symbol across planes. This symbol conversion is performed in plural (here, 8
The amount of information of the composite image of (color) symbols is determined by the large entropy color symbol, and the fact that the amount of image information of the color symbol varies greatly depending on the symbol is used. That is, the correlation with the adjacent pixel
Character images and mark images with low entropy
(The probability that the value is the same as that of the contact pixel is low.)
Low (the probability of the same value as the neighboring pixels is high)
No). Therefore, character image, mark image, solid image 1
And four image display elements of solid image 2,
If you rank them in order of higher entropy,
(1) Character image, (2) Mark image, (3) Solid image
1, (4) solid image 2. About these
The assigned symbol is assigned. Here, 3-bit encoding
Codes are assigned as symbols and in order of higher entropy
Is assigned to the code "001"
And assign “010” to the mark image,
Assign “100” and assign “110” to solid image 2.
"000" is assigned to the background color to be assigned. This
By doing so, the 3-bit code symbol is
The order of bit 2, bit 1 and bit 0 from the
And a binary image of bit 0 (bit plane 0)
Image) is an image of only a character image. Also, bit 1
Binary image (one bit plane image) is
The image 2 is the image obtained by taking the logical OR of the two images. More
The binary image of bit 2 (bit plane 2 image) is marked
The image is an image only. Therefore, bit plane 0
The entropy of the image is the character with the highest entropy
It is composed only of images, and the entropy of character images
Decided. One bit plane image is a mark image and a solid image.
Image 2 has an entropy of almost both
It becomes sum. Bit plane 2 image is solid image 1 and solid image
Image 2 has an entropy of almost both
It becomes sum. As a result, the entropy of the image to be compressed is
Is the sum of the entropy of the image display element as
You. Character image + mark image + 2 * (solid image 1 + solid image
Image 2) As can be seen from the above equation, the entropy should be low.
Image 1 and solid image 2 are doubled, but the entropy
Character images and mark images with large
No. As a result, it is composed of characters, marks, solid images, etc.
When realizing multi-color (multi-bit) documents,
According to the encoding symbol conversion of the embodiment, multi-color
The entropy afterwards is a binary image (the system of all image display elements).
Monochrome image when the symbol is expressed as 1/0)
Not much different from tropy. Therefore, multi-color
The amount of data after compression of the subsequent image is binary (monochrome)
It is the same as the data amount when the image is compressed. Although the amount of information of the original image of the color extended image is three times as large as that of the binary image, according to the compression apparatus using symbol conversion of the present invention, the amount of information after compression is 2 when the encoding means is the same. It approaches the amount of information after compression to the value image. The time required for encoding in the compression apparatus of the present invention is the same as the time required for encoding the (R, G, B) color symbol, because the number of planes is the same, that is, Since the numbers are the same, the time will be the same. The time required for symbol conversion can be ignored if realized by hardware logic, and is sufficiently small even in a lookup table format using the table described in the second embodiment. The size of the compression device of the present invention is R, G, B
In comparison with the color symbol compression apparatus described above, only a symbol conversion unit is added, and the size is small according to the above-described realization method. Here, a method of calculating entropy using a Markov model will be described. FIG. 9 is used for explanation. In the text, the Markov state m determined by the value of the reference pixel is quantified by Expression 1. M = a · 32 + b · 16 + c · 8 + d · 4 + e · 2 + f— (Equation 1) where a, b, c, d, e, and f are shown in FIG. Indicates the value (1/0) of the corresponding pixel. Consider a case where a certain coded pixel x is focused on a compression target image (a map in this embodiment). From the reference pixel value determined by this coded pixel,
, And the value (0/1) of the coded pixel x is checked at the same time. This is performed on all the pixels according to the encoding order 9, for example, and the occurrence frequency a0 [m] of 1 and the occurrence frequency a1 of 1 for all 64 Markov states
[M] is obtained. Further, the occurrence frequency b [m] of 64 Markov states is obtained. Accordingly, the occurrence probability Pa [1 / m] of 1 and the occurrence probability Pa of 0 in a certain Markov state m
[0 / m] is calculated by Expressions 2 and 3. Also, the Markov state occurrence probability Pb [m] is calculated by Expression 4. Pa [0 / m] = a0 [m] / (a1 [m] + a0 [m]) − (Equation 2) Pa [1 / m] = a1 [m] / (a1 [m] + a0 [m])-(Equation 3) Pb [m] = b [m] / T- (Equation 4) Here, T is the total number of pixels. From these occurrence probabilities, the entropy H [m] is calculated according to equation (5). [0037] Here, the bottom of log is 2. The method of calculating entropy is described in, for example, "Theory of Information and Code" published by Iwanami Shoten. Next, a document image created by a computer according to an embodiment of the present invention is color-extended to, for example, 3 bits.
An apparatus for compression will be described with reference to FIG . In this case as well, the information amount of the original image, which is tripled by the color expansion , can be made close to the information amount after compression of the image represented by the binary image by the first compression device. This embodiment will be specifically described with reference to FIG. Normally, in a multi-color document image, characters, line drawings, ruled lines, solid colors, and the like are assigned to each color (symbol).
The amount of information of a symbol (color) as an image varies greatly depending on the symbol because the statistical properties of the image are different. For example, characters have a large amount of information, and line drawings and solid colors are small. By utilizing this, symbol allocation can be determined according to the procedure described in the first embodiment. In the present embodiment, since the relationship between the color and the coded symbol is not always fixed, the correspondence between the color symbol and the coded symbol of the symbol converting means 1 (herein the text) before the coding process. This is called a symbol conversion table). For this purpose, the configuration of the symbol conversion means 1 is a look-up table method using, for example, a RAM or the like. The input values of the RAM table are color symbols, and the output values are coded symbols. The realization of this table is based on the color-coded symbol table information 8 in FIG.
Can be written in the RAM. After writing the symbol conversion table, the color symbols (R, G, B) of the image information source 2 are input to the symbol conversion means 1, and the coded symbols (D0, D1, D2)
Convert to The encoded symbol is encoded by the encoding means 3 to generate compressed data 7. The encoding means 3 uses the one shown in the first embodiment, for example. The compressed data 7 is
For example, as shown in FIG. 7, the recording format is converted into a recording format synthesized with the color-coded symbol information of the attribute information by the recording format conversion means 5 and is recorded in the recording device 4. Coded symbols include characters, lines, fonts,
It can also be interpreted as a symbol indicating a fill, a figure, a background, or the like. Therefore, as another effect of improving the compression ratio, a multicolor image of an arbitrary color can be reproduced by changing the symbol conversion table. For example, only characters can be changed to red. Also, a bitmap can be formed by extracting only an arbitrary symbol. For example, a bitmap can be formed only from figures in a document. If this is used, more intelligent document processing and output can be performed by converting the binary representation of the document into multi-valued data. (Embodiment 2) In this embodiment, the compression apparatus of the present invention is applied to an image read by a color scanner or the like. This embodiment will be described with reference to FIG. In the present embodiment, the image information source 2 receives R, G, B from a color scanner.
It is. Further, as shown in FIG. 4, image data from a color scanner may be stored in a frame buffer. Before performing the encoding process, preprocessing 6 is performed using the image information source 2. This is a process of obtaining the image information amount of the color symbol described in the first embodiment, and deciding the symbol assignment in accordance therewith. The pre-processing 6 may be the procedure described in the first embodiment, but it takes a long time to calculate, and therefore, this embodiment proposes another method. This will be described later. The symbol conversion means 1 determined by the preprocessing 6 can change the symbol conversion table because the statistical properties of the image to be encoded cannot be predicted (described in the second embodiment).
And After the completion of the preprocessing, that is, after writing the symbol conversion table, the encoding processing described below is performed. Based on the symbol conversion table (1),
Color symbols (R, G, B) from a color scanner or a frame buffer are encoded into coding symbols (D0, D
1, D2). The encoded symbol is encoded by encoding means 3
And the compressed data 7 is generated. The encoding means 3 is, for example, as shown in the first embodiment. As described in the second embodiment, the compressed data 7 is recorded in the recording format by the recording format conversion means 5 and recorded in the recording device 4. The construction method of the symbol conversion table proposed in this embodiment, that is, the processing contents of the preprocessing 6 will be described. Here, the image information amount of the symbol necessary for the preprocessing 6 is defined as the occurrence frequency of the symbol. That is, the coded pixels are swept in the dot order as shown in FIG. 9 (referred to as pre-scan in the text), and the frequency of occurrence of color symbols is counted. 1
After pre-scanning all images, the frequency of occurrence of each color symbol is defined as the image information amount. For the prescan, a color scanner may be driven for that purpose, or data accumulated in a frame buffer as shown in FIG. 4 may be used. Alternatively, a method may be used in which a plurality of representative points suitable for an image are determined, and only those points are sampled to count the frequency of occurrence of color symbols. When the occurrence frequency is known, symbol assignment is performed according to the following procedure. Procedure 1: Arrange in the order of occurrence frequency. In this embodiment, eight color symbols are ordered. Step 2: 000 is assigned to the background color symbol which usually occurs most frequently. Step 3: Excluding the background color,
001, 010,
Assign 100. The order is not specified in this embodiment. Step 4: 011, 101, 1
Assign 10 The order is not specified in this embodiment. Step 5: 111 is assigned to the symbol having the lowest occurrence frequency. Advantages of storing a multi-color image in the form of a bitmap include a reduction in calculation load for drawing for a computer, high-speed drawing, and sharing of data. By applying the compression device of the present invention to the bit-mapped multicolor (multi-valued) image, the following effects can be obtained. Although the information amount of the original image of the color extended image is twice (the number of planes) times that of the binary image, according to the compression apparatus using the symbol conversion of the present invention, the information amount after compression is If the coding method is the same, the amount of information of the compressed binary image can be approximated. Since the decoding / reproduction time of a compressed image is determined by the number of planes, that is, the number of coded pixels, it does not change in the apparatus of the present invention. Further, the processing time required for symbol conversion is almost negligible compared to the time required for encoding. The hardware scale of the symbol conversion means is as follows.
It is sufficiently smaller than the encoding means. More intelligent document image processing becomes possible.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram showing an example of a compression device according to the present invention. FIG. 2 is a diagram showing an example when the present invention is applied to a multi-color document image. FIG. 3 is a diagram showing an example when the present invention is applied to a color scanner. FIG. 4 is a diagram in which a frame buffer is provided as an image information source in FIG. 3; FIG. 5 is a diagram for explaining the concept of a coded symbol according to the present invention. FIG. 6 is a diagram showing an example of an encoding point order of the compression device of the present invention. FIG. 7 is a diagram showing an example of a recording format of compressed data by the compression device of the present invention. FIG. 8 is a view for explaining a conventional encoding method for image compression. FIG. 9 is a view for explaining an encoding method based on a Markov model. [Description of Codes] 1 ... Symbol conversion means 2 which is a main element of the present invention 2 ... Multi-color (multi-valued) image information source 3 to be compressed 3 ... Encoding means 4 for each bit plane ..Compressed data recording device 5... Means for converting compressed data into a recording format 6... Preprocessing means 7 for deciding symbol assignment in symbol conversion table 7... Compressed data 8. Information 9: encoding order, 10: Example of a Markov model model template 11: Pixel prediction means 12 using a Markov model is a dynamic arithmetic encoding means

Claims (1)

(57) [Claims] [Claim 1] Image display of characters, line drawings, solid images, etc.
Bit plane composed of elements and two or more layers
A device for compressing multi-valued images with
Symbol converting means for converting the display symbols indicating the image display element to the encoding symbols, the encoding symbols An image compression apparatus comprising a compression means for encoding for each bit plane, the symbol converting means Determining an n-bit encoded symbol from the n-bit display symbol indicating the image display element.
Having a conversion table for determining
Is a two-dimensional image of a binary image formed only with each image display element.
The statistical properties of an image are determined from the correlation with adjacent images.
Entropy for all image display elements
Calculate and rank each image element in descending order of entropy
In addition, among the n image display elements, the entropy is large.
For thresholds, each is assigned to a coded symbol with only one other bit and zero other bits , and the entropy is
For the smallest image display element, all n bits are assigned to one coded symbol, and
An image compression apparatus for allocating to n coded symbols of all 0 bits .