JP3320282B2 - Multi-value image processing device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for extracting bit data of the same bit order for each pixel from a monochrome multi-valued image composed of a plurality of bits per pixel, forming bit planes, and performing arithmetic coding on each bit plane. A multi-valued image processing apparatus that applies an arithmetic decoding process, which is the inverse process of the above-described arithmetic coding process, to the multi-valued image code data formed by the above, and reproduces the original image, For each color component, bit data having the same bit order is extracted for each pixel from a multi-valued color image composed of a plurality of bits per pixel to form a bit plane, and each color plane is formed by arithmetic code processing. A multi-valued image processing device that applies an arithmetic decoding process, which is a reverse process of the above-described arithmetic coding process, to the value image code data and reproduces the original image. On.

[0002]

2. Description of the Related Art In recent years, with the diversification of data communication and data processing, for example, as a document to be exchanged by a facsimile device or processed by a personal computer device, a monochrome multi-gradation comprising a plurality of bits per pixel is used. There is a demand for using an image or a color multi-tone image composed of a plurality of color components and one color component composed of a plurality of bits per pixel.

Here, as is well known, since the data amount of image data is very large, if the data is processed as it is, for example, the capacity of a storage device required for storage becomes enormous, Since the time required for data processing becomes excessively long, usually, data communication is performed or accumulated in a state where image data is encoded and compressed to reduce the data amount. In particular, since a black-and-white multi-tone image and a color multi-tone image have a large number of bits per pixel, an encoding method capable of encoding and compressing with higher efficiency is required.

For example, when encoding and compressing a black-and-white multi-tone image, an encoding method (so-called JPEG method) in which a plurality of bits per pixel is handled as it is.
A so-called bit plane in which bit data of the same bit order in a monochrome multi-tone image is extracted for each pixel to form a bit plane, and a binary encoding process is performed for each bit plane to encode and compress the original image. There are two types of encoding methods. In the latter encoding method, an appropriate binary image encoding method, for example, an MH method, an MR method, an MMR method, a JBIG method, or the like is adopted as a binary encoding process.

Here, in the former coding method described above,
Because it is a so-called lossy encoding method that partially reduces the information amount of the original image to the extent that the image quality is not impaired using human visual characteristics, the character image in the reproduced image is blurred, or if the encoding compression ratio is increased, There is a situation that the image quality of the reproduced image is extremely deteriorated, and in the latter encoding method, the encoding compression rate when encoding and compressing a monochrome multi-tone image is slightly inferior to the former encoding method. There are circumstances.

[0006]

By the way, in the case of decoding the encoded data and processing the reproduced image, in order for the characteristics of the output image to be appropriately represented by the output means,
The output image may be pre-processed depending on the output characteristics of the output unit and the type of the image.

The present invention has been made in view of the above circumstances, and provides a multi-valued image processing apparatus capable of obtaining a high-quality black-and-white multi-valued image or a color multi-valued image according to the type of image. It is intended to be.

[0008]

According to the present invention, a bit plane is formed by extracting bit data of the same bit order for each pixel from a black-and-white multi-valued image composed of a plurality of bits per pixel, and forming an arithmetic operation on each bit plane. In a multi-level image processing apparatus that applies an arithmetic decoding process, which is the inverse process of the above-described arithmetic coding process, to multi-level image code data formed by code processing and reproduces an original image, code data of each bit plane is reproduced. Arithmetic decoding means for applying the arithmetic decoding processing to form the original binary image data, and the symbol appearance probability in each state for each state determined at the time of decoding processing by the arithmetic decoding processing means. State statistical means for collecting and statistic data, and an image determining means for determining the type of the image to be processed based on the statistical result of the state probability statistical means. And image processing means for performing image processing on decoded multivalued image data obtained by the arithmetic decoding processing means for all bit planes based on the type of image determined by the image determination means. .

Also, from a multi-valued color image having a plurality of color components and a plurality of bits per pixel, for each color component, bit data having the same bit order is extracted for each pixel to form a bit plane. In a multi-valued image processing apparatus that applies an arithmetic decoding process, which is a reverse process of the above-described arithmetic coding process, to a color multi-valued image code data formed by performing an arithmetic coding process on a bit plane and reproduces an original image, Arithmetic decoding processing means for applying the arithmetic decoding processing to the bit plane code data to form the original binary image data; and states for each of the states determined at the time of decoding processing by the arithmetic decoding processing means. State probability statistic means for collecting and statistic symbol appearance probabilities at a time, and an image to be processed based on the statistical result of the state probability statistic means. An image determining unit of type determines,
Image processing means for performing image processing on decoded color multi-valued image data obtained by the arithmetic decoding processing means for all bit planes of all color components based on the image type determined by the image determination means. It is a thing.

[0010] The image determining means may determine whether the occurrence probability of the inferior symbol in a predetermined state corresponding to a feature of a character image appearing in a statistical result corresponding to a bit plane of a predetermined upper bit is smaller than a predetermined value. It is determined that the image to be processed is a character image, and when the inferior symbol appearance probability of the predetermined state, which appears in the statistical result corresponding to the bit plane of the predetermined upper bit, is equal to or more than a predetermined value, the It is determined that the image to be processed is a photographic image.

[0011] The image determining means may be arranged such that when the probability of occurrence of a less-probable symbol in a predetermined state corresponding to a feature of a character image appearing in a statistical result corresponding to a bit plane of the most significant bit is smaller than a predetermined value, When the image to be processed is determined to be a character image, and the occurrence probability of the inferior symbol in the above-mentioned predetermined state, which appears in the statistical result corresponding to the bit plane of the most significant bit, is equal to or more than a predetermined value, It is determined that the image is a photographic image.

[0012] When the appearance probability of the inferior symbol appearing in the statistical result corresponding to the bit plane of the most significant bit is biased to one of the states, the image to be processed is a character image. Is determined,
If the occurrence probability of the inferior symbol appearing in the statistical result corresponding to the bit plane of the most significant bit is not biased in any state, it is determined that the image to be processed is a photographic image.

[0013]

BRIEF DESCRIPTION OF THE DRAWINGS FIG.
Embodiments of the present invention will be described in detail.

First, the basics of the encoding system used in this embodiment will be described.

In this embodiment, when encoding and compressing a black-and-white multi-valued image, bit data having the same bit order is extracted for each pixel to form a bit plane, and a binary encoding process is performed for each bit plane. In this case, a so-called bit plane coding method for coding and compressing an original image is adopted.

For example, when an 8-bit multi-valued image per pixel is encoded by the bit plane method, first, as shown in FIG.
The bit (MSB) is extracted to form a bit plane BP8, and the seventh bit is extracted to extract a bit plane BP7.
Are formed in the same manner up to the first bit (LSB) bit plane BP1.

The bit planes BP8 to BP1 of each bit formed as described above can be handled as binary image data. In the present embodiment, a so-called JBIG method is used as an encoding method for compressing binary image data with high efficiency.

In the coding system of the JBIG system, an image is regarded as a Markov model, and coding is performed with its state recognized. That is, using the model template, a plurality of reference pixels around the target pixel to be encoded are extracted, and the state of the target pixel is predicted for each bit pattern of the extracted reference pixels. The binary image data is encoded and compressed by performing QM-coder encoding based on the prediction probability.

Here, the arithmetic code which is the basis of the QM-coder coding method will be described. Arithmetic code means that a corresponding section (binary decimal [0.0... 0, 0.1... 1)) on a number line from 0 to 1 ([0, 1)) according to the occurrence probability of each symbol. The target symbol sequence is assigned to the corresponding subsections, and the coordinates of the points included in the section obtained by repeating the division recursively are distinguished at least from other sections. It is represented by a binary decimal number that can be used as a code as it is.

For example, in binary arithmetic coding for the coded symbol sequence 0100, coding as shown in FIG. 2 is performed.

In FIG. 1, for example, when encoding the first symbol, the entire interval is divided into A (0) and A (1) according to the ratio of the occurrence probabilities of the 0 and 1 symbols. A (0) is selected. Next, when encoding the second symbol, A (0) is further divided by the occurrence probability ratio of both symbols in that state, and A (01) is selected as a section corresponding to the generated symbol sequence. Coding is advanced by repeating such division and selection processing.

This arithmetic code is a so-called non-block code, and in particular, the appearance probability of a symbol (binary (black and white) state of data) changes according to the state number (context) as in the above-described predictive coding method. It is suitable for encoding in such a case, and requires less hardware such as an encoder scale and a required memory amount than a run-length code such as an MH code, and higher efficiency can be expected. There is an advantage that encoding is easy.

The above-described QM-coder implements this arithmetic coding using less hardware resources and enables higher-speed processing. Normally, image data is not directly subjected to arithmetic coding, but pre-processing is performed by predictive coding to determine whether each pixel of the image data is a lesser symbol or a more significant symbol. Here, the inferior symbol represents a symbol having a lower occurrence probability in the context at that time, and the dominant symbol represents a symbol having a higher occurrence probability in the context at that time. Therefore, whether the inferior symbol or the superior symbol is a white pixel or a black pixel is statistically determined according to the context.

Next, the encoding / decoding algorithm of QM-coder will be described.

At the beginning of encoding, a number line from 0 to less than 1 is considered. Assuming that the interval between the number lines is A, and the probability of occurrence of inferior symbols corresponding to the Markov state (state 0 in the initial state) at that time is Qe, the area after division is represented by the following equation (I).

A = A × (1−Qe) (a: size of region of superior symbol) b = A × Qe (b: size of region of inferior symbol) (I)

Here, if the first symbol is 0 (dominant symbol), a is set as a new A, and if it is 1 (inferior symbol), b is set as a new A and further division is performed. The encoding table stores a set of a relationship between a Markov state number and a probability Qe, information on a state transition when a renormalization process occurs (described later), and the like.

However, performing the multiplication operation at the time of encoding is disadvantageous in terms of the apparatus scale and the operation speed. Further, when encoding an infinite-length symbol sequence, an infinite-length operation register is required to operate a reduced area.

Therefore, the QM-coder operates so that the new A after the division always has a size of 0.75 or more and less than 1.5, so that A can be approximated to 1 and the above-mentioned equation (I ) Is approximated by the following equation (II).

A = A−Qe b = Qe (II)

In accordance with this approximation, when the value of A becomes less than 0.75, renormalization processing in which A is bit-shifted to the left and enlarged until A becomes a value of 0.75 or more and less than 1.5. Perform By this renormalization processing, an operation requiring multiplication can be realized by subtraction, and an operation can be performed using a finite length register. When the renormalization process is performed, the state changes to the next Markov state according to the Markov state at that time and the distinction between the superior / inferior of the symbol to be processed.

The encoding process of the superior symbol is performed, for example, as shown in FIG. 3A, and the encoding process of the inferior symbol is performed, for example, as shown in FIG. Here, C represents a code, and its initial value is 0.

That is, when encoding the dominant symbol, the code C is not changed (process 101), and the value of the area A is updated to a value smaller by the probability Qe corresponding to the Markov state at that time (process 102). At this time, it is checked whether or not the value of the area A is smaller than 0.75 (determination 10
3) When the result of the judgment 103 is YES, the value of the area A and the code C are renormalized and the state is changed (processing 104), and the processing of one dominant symbol is ended. When the result of the judgment 103 is NO,
The process 104 is not performed, and the Markov state at that time is maintained.

When coding the inferior symbol, the value of code C is increased by (A-Qe) (step 20).
1) Update the value of the area A to the value of the probability Qe (Process 20)
2) The region A and the code C are re-normalized and the state is changed (process 203).

The code C generated at this time matches the binary decimal value indicating the lowermost part of the area. In the re-normalization process, the code C is shifted to the left by the same number of digits as the area A and enlarged, and the value of the code C exceeding 1 is output as code data.

FIG. 3C shows an example of the decoding process.

First, it is checked whether or not the code C is smaller than the value (A-Qe) (decision 301). If the result of the decision 301 is YES, the target pixel to be decoded is set as the dominant symbol. Judgment is made (step 302), and the value of the area A is updated to a value reduced by the probability Qe corresponding to the current Markov state (step 303). Then, it is checked whether or not the updated value of the area A has become smaller than 0.75 (determination 304). When the result of the determination 304 is YES, the area A and the code C are renormalized and the Markov state is changed. After the transition (process 305), the decoding process of this one symbol ends. When the result of the determination 304 is NO, the process 305 is not performed, and the Markov state at that time is maintained.

If the result of determination 301 is NO, the pixel of interest is determined to be the inferior symbol (step 30).
6) Update the value of code C to a value smaller by (A-Qe) (process 307), update region A to the value of probability Qe (process 308), and renormalize region A and code C. Together with the Markov state (process 309).
The decoding process for one symbol is completed.

As described above, when encoding the QM-coder, the code C does not change when a superior symbol appears, and there is little possibility that renormalization processing will be performed. The re-normalization process is performed and code data is formed.

Therefore, if the predictive coding process, which is the pre-processing, is performed so that the appearance frequency of the inferior symbol becomes small, the coding efficiency is improved and the processing speed is also improved.

FIG. 4 shows an example of a template used for the predictive encoding process. This template is JB
This is a so-called JBIG default three-line template used as a basis (default) in the IG system. In this template, 10 pixels around the target pixel are extracted as reference pixels. In this case, 1024 states (contexts) according to the bit pattern of the reference pixel are used.
Is determined.

FIG. 5 shows an example of a multi-level coding apparatus for coding and compressing multi-level image data by applying the JBIG method.

In the figure, multi-valued image data PX having, for example, an 8-bit width is input through an image input unit 1.
It is added to the bit plane development unit 2. The bit plane development unit 2 develops multi-valued image data into bit planes, and the data of each bit plane is
It is stored in the bit plane memory 3.

The data reference unit 4 reads the data of the target pixel to be coded from the bit plane memory 3 and outputs it to the arithmetic code engine 5 as target pixel data DX. The template is applied to read out the data of the plurality of reference pixels from the bit plane memory 3, and the reference pixel data DR
Is output to the probability evaluator 6.

The probability evaluator 6 determines the context based on the input reference pixel data DR, and for each context, the probability estimation value of the inferior symbol (or superior symbol) and the inferior symbol (or superior symbol). Is output to the arithmetic code engine 5.

The arithmetic coding engine 5 executes the above-described coding processing based on the input pixel data of interest DX, the estimated value of the inferior symbol (or superior symbol), and the type of the inferior symbol (or superior symbol). And
The code data CX obtained as a result is output to the code generator 8. Further, when the re-normalization process is performed in the encoding process, the arithmetic encoding engine 5 notifies the probability evaluator 6 to that effect.

Accordingly, the probability evaluator 6 changes the context in synchronization with the execution of the re-normalization process, and updates the probability evaluation state. In addition, the probability evaluator 6 calculates a Markov state value (probability estimation value index value: 7 bits) and a 1-bit symbol indicating that the dominant symbol at that time is either white or black for each context, for a total of 8 bits. Data is stored.

The arithmetic code engine 5 and the probability evaluator 6 constitute an arithmetic coder 7.

Here, the template used by the data reference unit 4 uses the same JBIG default three line template as shown in FIG. 4, so that ten reference pixels A to J are extracted for the target pixel X. The extracted 10-bit data of the reference pixel is output to the probability evaluator 6 as reference pixel data DR.

Also, in this case, the probability evaluator 6
For each of the 024 contexts, 8-bit data (probability estimated value index and value of dominant symbol) is stored.

The code creating section 8 forms multi-level code data CD corresponding to the code data CX added from the arithmetic encoder 7, and outputs the multi-level code data CD to the next stage device.

Here, an example of the multi-level code data CD formed by the code generation section 8 is shown in FIG.

The multi-level code data CD starts with bit precision information (n; "8" in this case) corresponding to the number of bits of the original image data PX, and then moves from the most significant bit plane to the least significant bit plane. Bit plane code data corresponding to each bit plane is sequentially arranged.

FIG. 7 shows an example of a multi-level decoding processor for decoding multi-level code data CD according to an embodiment of the present invention.

In FIG. 1, multi-level code data CD input through a data input unit 10 is applied to an arithmetic decoding engine 12 of an arithmetic decoder 11, and is decoded and output from the arithmetic decoding engine 12. Image data CDx
Are added to the bit plane data creation unit 13.

The bit plane data creation unit 13 stores the image data CDx added from the arithmetic decoder 11 in the bit plane memory 14 of the corresponding bit plane, and stores the reference pixel data designated by the data reference unit 15 in the bit plane. The data is read from the plane memory 14 and output to the data reference unit 15.

The data reference unit 15 applies a predetermined template to the pixel of interest to be decoded, specifies the address of the required reference pixel to the bit plane data generation unit 13, Further, the reference pixel data is received, and the received reference pixel data is output to the probability evaluator 16 of the arithmetic decoder 11.

The probability estimator 16 determines the context based on the input reference pixel data, and for each context, the probability estimation value of the inferior symbol (or superior symbol) and the probability of the inferior symbol (or superior symbol). The type is output to the arithmetic decoding engine 12 and state number data DC representing the state number of the context obtained at that time is represented by the state probability statistic unit 17.
Output to

The arithmetic decoding engine 12 inputs the target pixel data, the estimated value of the inferior symbol (or superior symbol), and the inferior symbol (or superior symbol).
And outputs the resulting image data CDx to the bit-plane data creation unit 13 based on the type of the symbol, and determines whether the pixel of interest at that time is a inferior symbol or a superior symbol Is output to the state probability statistical unit 17. Further, when the renormalization processing is performed in the decoding processing, the arithmetic decoding engine 12 notifies the probability evaluator 16 of the renormalization processing.

Accordingly, the probability evaluator 16 changes the context in synchronization with the execution of the re-normalization process, and updates the probability evaluation state. Further, the probability evaluator 16 calculates a Markov state value (a value of the probability estimation value index; 7 bits) and a 1-bit symbol indicating that the dominant symbol at that time is either white or black for each context.
Bit data is stored.

Here, the data reference unit 15 uses a JBIG default three line template as shown in FIG. Therefore, for the target pixel X, the reference pixels A to
The ten reference pixels of J are extracted, and 10-bit data of the extracted reference pixels is output to the probability evaluator 16 as reference pixel data.

In this case, the probability evaluator 16
For each of the 1024 contexts, 8-bit data (probability estimated value index and value of dominant symbol) is stored.

The state probability statistic unit 17 counts the number of appearances of the inferior symbol and the superior symbol for each context for a predetermined bit plane (for example, the most significant bit plane), and calculates the number of appearances of the inferior symbol and the superior symbol. The sum is calculated, the number of occurrences of the inferior symbol is divided by the value of the sum, and the probability of appearance of the inferior symbol for each context is calculated.
The calculation result is output to the image determination unit 18 for each bit plane as the state probability statistical data SP.

The image determination unit 18 determines whether the image decoded and reproduced is a character image or a photographic image based on the state probability statistical data SP input from the state probability statistical unit 17. The determination result is provided to two linked switches 19 and 20 as image determination data KP.

The switch 19 is provided in the bit plane memory 14
The image data at the same pixel position is respectively taken out from all the bit planes and combined to form multi-valued image data XP, and the multi-valued image data XP is used when the value of the image determination data KP represents a photographic image. Is output to the photo image processing unit 21, and is output to the character image processing unit 22 when the value of the image determination data KP indicates a character image.

When the value of the image determination data KP indicates a photographic image, the switch 20 switches to the photographic image processing section 21.
And the image determination data KP
If the value represents a character image, the output data of the character image processing unit 22 is selected, and the selected data is output to the next-stage device as multi-valued image data XPa.

Here, the photographic image processing section 21 operates as shown in FIG.
As shown in (b), image processing (γ correction processing) such that the output image density has a linear relationship with the input document density, or the density of the target pixel is smoothed using surrounding pixels. Is what you do. Thereby, the switches 19 and 20
When the photographic image processing unit 21 is selected by
The multi-valued image data XP is output as multi-valued image data XPa converted into a state in which the halftone state is well represented by the photographic image processing unit 21, and as a result, the reproduced image quality of the photographic image is improved.

Further, the character image processing section 22 operates as shown in FIG.
As shown in (1), the image density to be output with respect to the input document image has a non-linear relationship, which is caused by image processing (γ correction processing) for emphasizing an edge portion appearing in a character image, pixel structure of a reading unit, and the like. To perform smoothing processing or the like for eliminating jaggedness appearing at the edge of the character image. Thereby, the character image processing unit 22 is switched by the switches 19 and 20.
Is selected, the multi-valued image data XP is output as multi-valued image data XPa converted by the character image processing unit 22 into a state in which the characteristics of the character image are well expressed. As a result, the character image Good reproduction image quality.

Therefore, when the image determination unit 18 determines that the image of the multivalued image data XP to be processed is a photographic image, the image determination data KP corresponding to the photographic image is output. 19 and switch 20
Selects the photographic image processing unit 21, the multi-valued image data XP output from the bit plane memory 14 is subjected to the above-described image processing by the photographic image processing unit 21, Since the image is output as XPa, the reproduction image quality of the photographic image is improved.

When the image of the multivalued image data XP to be processed is determined to be a character image by the image determination section 18, image determination data KP corresponding to the character image is output. Since the multivalued image data XP output from the bit plane memory 14 is subjected to the above-described image processing by the character image processing unit 22 and the switching unit 2
Since it is output as multi-valued image data XPa via 0, the reproduction quality of the character image is improved.

As described above, in the present embodiment, the optimum image processing is performed for each of the reproduced image in the case of the photographic image and the case of the character image, and the image quality is good regardless of the type of the reproduced image. It becomes something.

According to an experiment conducted by the inventor of the present invention, a method of determining whether an original image is a character image or a photographic image based on multi-valued image data has been described. Since the contents of the images are largely divided, it has been confirmed that the occurrence probability of the inferior symbol becomes extremely small for one or more specific state numbers representing them. Here, the specific state number is a state number corresponding to a bit pattern representing a vertical line or a horizontal line.

Accordingly, the average value of the appearance probabilities of the inferior symbols for the specific state number is calculated, and if the average value is larger than the predetermined value, it can be determined that the original image is a character image. If the average value is equal to or greater than a predetermined value, it can be determined that the original image is a photographic image.

FIG. 9 shows another example of the processing of the image judging section 18 when such a judging method is applied.

First, the state probability statistical data SP for one bit plane output from the state probability statistical unit 17 is input (processing 401), and the occurrence probability of the inferior symbol is extracted for each of the specific state numbers described above (processing 401). 402), and calculate the average value Pm of the extracted appearance probabilities (process 40).
3) It is checked whether or not the value of the average value Pm is larger than a predetermined value KA (determination 404).

If the result of determination 404 is YES, it is determined that the original image at this time is a character image (process 405), and the image determination data having contents that cause the switches 19 and 20 to select the character image processing unit 22 Output KP (Process 4
06).

If the result of determination 404 is NO, it is determined that the original image at this time is a photographic image (step 407), and the flow shifts to step 406. It outputs image determination data KP having contents that cause the image processing unit 21 to select.

As described above, in the present embodiment, it is possible to improve the quality of the reproduced image according to the type of the original image.

In the above-described embodiment, a case has been described in which multi-valued image data is processed when a monochrome image is read as multi-valued image data. However, in the present invention, a color image is read as multi-valued image data. The same can be applied to the case.

FIG. 10 shows an example of a color multi-level image encoding apparatus according to still another embodiment of the present invention. In this case, the color multi-tone image to be encoded is
It has color components of three primary colors RGB, and each color component has a bit depth of 8 bits per pixel. In this figure, the same parts as those in FIG. 5 and corresponding parts are denoted by the same reference numerals or related reference numerals.

In the figure, 24-bit color multi-tone image data P input through a color image input unit 25 is shown.
Xc is added to the color component separation unit 26. The color component separation section 26 converts the color multi-tone image data PXc into 8-bit R component R multi-tone image data PXr, 8-bit G component G multi-tone image data PXg, and 8-bit B
The component B is decomposed into B multi-gradation image data PXb, and R multi-gradation image data PXr and G multi-gradation image data P
Xg and B multi-tone image data PXb are added to the bit plane developing units 2r, 2g, and 2b, respectively.

The bit plane developing section 2r develops the R multi-valued image data PXr into bit planes.
The data of each bit plane is stored in the bit plane memory 3r.

The bit plane developing unit 2g develops the G multi-valued image data PXg into bit planes.
The data of each bit plane is stored in the bit plane memory 3g.

The bit plane developing unit 2b develops the B multi-valued image data PXb into bit planes.
The data of each bit plane is stored in the bit plane memory 3b.

The data reference section 4r reads the data of the target pixel to be coded from the bit plane memory 3r, and outputs the read data as the target pixel data DXr by the arithmetic encoder 9r.
, A predetermined template is applied to the pixel of interest at that time, data of a plurality of reference pixels is read from the bit plane memory 3r, and the arithmetic operation is performed as reference pixel data DRr. The data is output to a probability evaluator (not shown; see FIG. 5) of the encoder 9r.

The data reference section 4g reads the data of the target pixel to be coded from the bit plane memory 3g, and as the target pixel data DXg, outputs the arithmetic coder 9g.
, A predetermined template is applied to the pixel of interest at that time, data of a plurality of reference pixels is read out from the bit plane memory 3g, and the arithmetic operation is performed as reference pixel data DRg. The data is output to a probability evaluator (not shown; see FIG. 5) of the encoder 9g.

The data reference section 4b reads the data of the target pixel to be coded from the bit plane memory 3b, and outputs the data as the target pixel data DXb by the arithmetic encoder 9b.
, A predetermined template is applied to the pixel of interest at that time, data of a plurality of reference pixels are read from the bit plane memory 3b, and the arithmetic operation is performed as reference pixel data DRb. The data is output to a probability evaluator (not shown; see FIG. 5) of the encoder 9b.

Here, the template used by the data reference sections 4r, 4g, and 4b uses the same JBIG default three-line template as shown in FIG. The reference pixel is extracted, and the extracted 10-bit data of the reference pixel is output to the arithmetic encoders 9r, 9g, 9b as reference pixel data DRr, DRg, DRb, respectively.

The arithmetic coder 9r executes the above-described arithmetic coding process based on the input pixel data of interest DXr and reference pixel data DRr, and obtains the coded data as R component code data CXr. 27.

The arithmetic coder 9g executes the above-described arithmetic coding process based on the input target pixel data DXg and the reference pixel data DRg, and obtains the obtained code data as G component code data CXg. 27.

The arithmetic coder 9b executes the above-described arithmetic coding process based on the input pixel data of interest DXb and reference pixel data DRb, and obtains the code data obtained thereby as B component code data CXb. 27.

The code generator 27 includes arithmetic coder 9r, 9
code data CXr, CXg, C added from g, 9b
It forms color multi-level code data CDc corresponding to Xb, and outputs the color multi-level code data CDc to the next stage device.

Here, an example of the color multi-level code data CDc formed by the code generation section 22 is shown in FIGS.

As shown in FIG. 11A, the color multi-level code data CDc has bit precision information (n; "8" in this case) corresponding to the number of bits of the original image data PXc at the beginning, and Following header information indicating the number of components (in this case, “3”), code data of each color component is sequentially arranged.

The code data of each color component is
As shown in FIG. 2B, bit plane code data corresponding to each bit plane is sequentially arranged from the most significant bit plane to the lower bit plane.

FIG. 12 shows an example of a color multi-level image decoding apparatus for decoding color multi-level code data CDc.

In the figure, color multi-level code data CD
c is added to the code distribution unit 30, and the code distribution unit 30
The multi-level code data CDr of the R component, the multi-level code data CDg of the G component, and the multi-level code data CDb of the B component included in the color multi-level code data CDc are extracted, and the multi-level code data CDr, CDr, CDg and CDg are output to the arithmetic decoders 11r, 11g, and 11b in the bit plane order from the most significant bit plane to the least significant bit plane. These arithmetic decoders 11r, 11
The decoded image data CCr, CCg, and CCb respectively output from g and 11b are added to the bit plane data generators 13r, 13g, and 13b, respectively.

Bit plane data creation units 13r, 13
g, 13b are image data CCr, CCg, CCb added from the arithmetic decoders 11r, 11g, 11b, respectively.
To the bit plane memory 1 of the corresponding bit plane.
4r, 14g, and 14b, respectively, and stores the reference pixel data specified by the data reference units 15r, 15g, and 15b in the bit plane memory 1 respectively.
4r, 14g, and 14b, and read the data reference unit 1
5r, 15g, and 15b.

The data reference sections 15r, 15g, 15b
A predetermined template is applied to the target pixel to be decoded, and the address of the required reference pixel is specified to each of the bit plane data generation units 13r, 13g, and 13b, and the bit plane data generation units 13r and 13b are designated.
g, 13b, and receives the reference pixel data, and converts the received reference pixel data into arithmetic decoders 11r, 1r.
1g and 11b are output to probability estimators (not shown; see FIG. 7).

Here, the data reference section 15 uses a JBIG default three-line template as shown in FIG.

In the arithmetic decoder 11r, the probability estimator determines the context based on the input reference pixel data, and for each context, the probability estimation value of the inferior symbol (or superior symbol) and the inferior symbol ( Or, the type of the superior symbol is output to the arithmetic code engine, and the state number data DC representing the state number of the context obtained at that time is output to the state probability statistic unit 17.

In the arithmetic decoder 11r, the arithmetic decoding engine performs the above-described decoding based on the input target pixel data, the estimated value of the inferior symbol (or superior symbol), and the type of the inferior symbol (or superior symbol). And outputs the image data CCr obtained as a result to the bit plane data generation unit 13 and also outputs the pixel type data DP indicating whether the pixel of interest at that time is the inferior symbol or the superior symbol to the state probability Output to the statistics unit 17. Further, when the renormalization process is performed in the decoding process, the arithmetic decoding engine notifies the probability evaluator of the fact.

Accordingly, the probability evaluator performs a context transition and updates the probability evaluation state in synchronization with the execution of the renormalization process. Further, the probability estimator calculates a Markov state value (a value of a probability estimation value index; 7 bits) and a 1-bit data indicating that the dominant symbol at that time is either white or black for each context, for a total of 8 bits. I remember.

Further, the state probability statistic unit 17 counts the number of appearances of the inferior symbol and the superior symbol for each context for a predetermined bit plane (for example, the most significant bit plane), and calculates the number of appearances of the inferior symbol and the superior symbol. The sum is calculated, the number of occurrences of the inferior symbol is divided by the value of the sum, and the probability of appearance of the inferior symbol for each context is calculated.
The calculation result is output to the image determination unit 18 for each bit plane as the state probability statistical data SP.

The image determining unit 18 determines whether the image decoded and reproduced is a character image or a photographic image based on the state probability statistical data SP input from the state probability statistical unit 17. The determination result is provided to two linked switches 19 and 20 as image determination data KP.

The image data synthesizing section 31 stores the bit plane image data CCr of the eighth bit plane to the first bit plane stored in the bit plane memory 14r.
Eighth bit plane to first bit plane bit plane image data CCg stored in bit plane memory 14g and eighth bit plane to first bit plane bit plane image data stored in bit plane memory 14b For CCb, image data of each pixel is sequentially extracted for each color component, and 8-bit data of three color components at the same pixel position are combined to form color multivalued image data XC. The image data XC is applied to the switch 19.

The switch 19 is provided with a color multi-valued image data CX.
Is output to the photo image processing unit 21 when the value of the image determination data KP indicates a photographic image, and is output to the character image processing unit 22 when the value of the image determination data KP indicates a text image. I do.

When the value of the image determination data KP indicates a photographic image, the switch 20 switches to the photographic image processing section 21.
And the image determination data KP
If the value represents a character image, the output data of the character image processing unit 22 is selected, and the selected data is
The data is output to the next-stage device as color multivalued image data CXa.

Therefore, when the image determination unit 18 determines that the image of the multivalued image data XP to be processed is a photographic image, the image determination data KP corresponding to the photographic image is output. 19 and switch 20
Selects the photographic image processing unit 21, the color multi-valued image data XC output from the image data synthesizing unit 31 is processed by the photographic image processing unit 21 in the image processing (γ
Correction processing and smoothing processing) are performed and output as color multi-valued image data XPa via the switch 20, so that the reproduced image quality of the photographic image is improved.

When the image of the multivalued image data XP to be processed is determined to be a character image by the image determination section 18, image determination data KP corresponding to the character image is output. And the switch 20 selects the character image processing unit 22, so that the image data combining unit 31
The color multi-valued image data XC output from the above is subjected to the above-described image processing by the character image processing unit 22 and is output as the color multi-valued image data XCa via the switch 20. Become good.

In this case, the judgment process of the image judgment unit 18 is as follows.
The same processing as in FIG. 9 can be applied.

As described above, according to the present embodiment, it is possible to improve the image quality of the color reproduced image according to the image type of the original image.

In the above-described embodiment, the JBIG default three-line template is used for encoding at the time of arithmetic coding. However, any other appropriate template (for example, a floating reference pixel is moved to an optimum state according to image content) Adaptive template, etc.).

[0114]

As described above, according to the present invention,
When reproducing a black-and-white multi-valued image from data encoded and compressed by the arithmetic coding process, image processing according to the type of the image is applied, so that the effect of improving the image quality of the reproduced image is obtained.

Further, when reproducing a color multi-valued image from data coded and compressed by arithmetic coding, image processing according to the type of the image is applied, so that the image quality of the reproduced color image can be improved. The effect that can be obtained.

Further, since the type of the reproduced image is determined based on the statistical information of the inferior symbol appearance probability obtained at the time of the arithmetic decoding process, the image type can be reliably determined. In this case, it is possible to prevent an inconvenience such as applying image processing for a character image.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram when a multi-valued image is encoded by a bit plane method.

FIG. 2 is a schematic diagram for explaining an example of binary arithmetic coding for a coded symbol sequence “0100”.

FIG. 3 is a flowchart for explaining an example of a QM-coder encoding / decoding process.

FIG. 4 is a schematic diagram showing an example of a template used for predictive encoding processing.

FIG. 5 is a block diagram showing an example of a multi-level encoding device according to one embodiment of the present invention.

FIG. 6 is a schematic diagram showing an example of multi-level code data.

FIG. 7 is a flowchart illustrating an example of processing of an encoding determination unit.

FIG. 8 is a graph for explaining a part of the processing (γ correction) of the photographic image processing unit and the character image processing unit.

FIG. 9 is a block diagram illustrating an example of a multi-level decoding device.

FIG. 10 is a block diagram showing an example of a color multi-level image encoding device according to another embodiment of the present invention.

FIG. 11 is a schematic diagram showing an example of color multi-level code data.

FIG. 12 is a block diagram illustrating an example of a color multi-level image decoding device that decodes color multi-level code data.

[Explanation of symbols]

11, 11r, 11g, 11b Arithmetic decoder 12 Arithmetic decoding engine 13, 13r, 13g, 13b Bit plane data creation unit 14, 14r, 14g, 14b Bit plane memory 15, 15r, 15g, 15b Data reference unit 16 Probability evaluator 17 State Probability Statistical Unit 18 Image Judgment Unit 19, 20 Switch 21 Photo Image Processing Unit 22 Character Image Processing Unit 30 Code Distribution Unit 31 Image Data Synthesis Unit

Claims (5)

(57) [Claims] 1. A bit plane is formed by extracting bit data of the same bit order for each pixel from a black-and-white multi-valued image composed of a plurality of bits per pixel, forming a bit plane, and performing arithmetic code processing on each bit plane. In a multi-level image processing device that applies an arithmetic decoding process, which is a reverse process of the above arithmetic coding process, to the value image code data and reproduces the original image, the arithmetic decoding process is applied to the code data of each bit plane. Arithmetic decoding means for forming the original binary image data, and a state probability for collecting and statistic symbol appearance probabilities in each state for each state determined during the decoding processing of the arithmetic decoding processing means. Statistical means; image determining means for determining a type of an image to be processed based on statistical results of the state probability statistical means; and image determining means A multi-valued image processing apparatus comprising: image processing means for performing image processing on decoded multi-valued image data formed and formed by the arithmetic decoding processing means for all bit planes based on the determined image type. . 2. A bit plane is formed by extracting bit data having the same bit order for each color component for each pixel from a color multi-valued image including a plurality of color components and a plurality of bits per pixel. A multi-valued image processing device that applies an arithmetic decoding process, which is a reverse process of the above-described arithmetic coding process, to a color multi-valued image code data formed by performing an arithmetic coding process on a bit plane, and reproduces an original image. Arithmetic decoding means for applying the arithmetic decoding process to the bit plane code data to form the original binary image data; and each state determined at the time of the decoding processing by the arithmetic decoding processing means. State probability statistical means for collecting and statistic symbol appearance probabilities in the image, and based on the statistical result of the state probability statistical means, Image determination means for determining a type; and decoded color multi-valued image data obtained by the arithmetic decoding processing means for all bit planes of all color components based on the image type determined by the image determination means. A multi-valued image processing apparatus comprising an image processing means for performing image processing on. 3. The image determining means according to claim 1, wherein said inferior symbol appearance probability in a predetermined state corresponding to a characteristic of a character image appearing in a statistical result corresponding to a bit plane of a predetermined upper bit is smaller than a predetermined value. It is determined that the image to be processed is a character image, and when the inferior symbol appearance probability of the predetermined state, which appears in the statistical result corresponding to the bit plane of the predetermined upper bit, is equal to or more than a predetermined value, the 3. The multi-value image processing apparatus according to claim 1, wherein the image to be processed is determined to be a photographic image. 4. The image determining means according to claim 1, wherein, when the occurrence probability of the inferior symbol in a predetermined state corresponding to the characteristic of the character image appearing in the statistical result corresponding to the bit plane of the most significant bit is smaller than a predetermined value, When it is determined that the image to be processed is a character image, and the occurrence probability of the inferior symbol in the above-mentioned predetermined state, which appears in the statistical result corresponding to the bit plane of the most significant bit, is equal to or more than a predetermined value, 3. The multi-value image processing apparatus according to claim 1, wherein the image is determined to be a photographic image. 5. The image determining means, when the occurrence probability of a less-probable symbol appearing in the statistical result corresponding to the bit plane of the most significant bit is biased to one of the states, the image to be processed is a character image. When the occurrence probability of the inferior symbol appearing in the statistical result corresponding to the bit plane of the most significant bit is not biased in any state, it is determined that the image to be processed is a photographic image. Claim 1 or Claim 2
The multivalued image processing apparatus according to the above.