JP3359214B2 - Multi-level image coding 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 multi-level image coding apparatus, and more particularly, to a prediction error obtained by subtracting a predicted value of a target pixel predicted from a level of a peripheral pixel from an actual pixel value of the target pixel. The present invention relates to a multi-level image encoding device that arithmetically encodes a value.

[0002]

2. Description of the Related Art Conventionally, an image handled by an image handling apparatus such as a facsimile apparatus or a personal computer is mainly a binary image represented by one bit per pixel due to a limitation of its communication capacity and storage capacity. Met.

However, if the original image to be converted into a binary image is a character document or the like, the loss of information after the binarization is small. When converted to, the loss of information was remarkable, and the image quality was inevitably degraded.

On the other hand, with the recent advance in technology, the processing capability of devices that handle images, such as facsimile machines and personal computers, has improved, and the quality of images to be handled has been increasing.

For this reason, a black-and-white multi-value (multi-tone) image in which a plurality of bits are represented per pixel, or a multi-valued image in which each color component is represented by a plurality of bits per pixel. Handling of a multi-valued color image or the like configured as a composition is relatively easy.

However, as compared with a binary image represented by one bit per pixel, the multi-valued image has a higher number of bits per pixel and the number of color components, and the higher the image quality, the higher the data quality. The amount increases. For example, when comparing the data amount of a black-and-white binary image and the data amount of a black-and-white 256-gradation image, the number of bits required for one pixel is 1 bit for the former, whereas the latter requires 8 bits. The data amount becomes eight times as large as the former.

[0007] Therefore, even if the processing capability of an apparatus for handling multi-valued images is improved and a large amount of image data can be processed at high speed, the transmission time of the multi-valued image data is limited due to the limitation of communication capacity and storage capacity. There is a problem in that the storage device for storing multivalued image data becomes excessively large and the storage device for storing multivalued image data becomes enormous.

As one method for matching the large data amount of the multi-valued image with the restriction on the communication capacity and the storage capacity, a multi-tone image is efficiently encoded and the encoded data is transmitted. For example, there is a method of memorizing.

[0009] ISO (International Organization for Standardization) is representative of an efficient coding method for multi-tone images (including color images).
And the ITU-T (formerly CCITT) standardized JPEG system.

[0010] The JPEG system is divided into two encoding systems. The first system is a basic DCT (Discrete).
e Cosine Transform), and the second method is an optional two-dimensional space DPCM (Differential PCM).
Is a DPCM method.

The DCT method is a method in which image information is converted into frequency information by using discrete cosine transform, quantized, and then encoded. This is a coding method (called a lossy coding method) that performs coding by partially reducing the amount of information.

The DPCM method is a method of predicting a target pixel level from surrounding pixels and coding a prediction error thereof. The coding method (lossless coding method and lossless coding method) performs coding without impairing the information amount of an original image. Called).

Comparing these two methods, if encoding is performed with emphasis on the compression ratio, it is an irreversible encoding method in which the original image cannot be completely reproduced in general because it involves quantization. It is preferable to use a DCT method having a large partial compression ratio. However, if encoding is performed with emphasis on information preservation, it is preferable to use the DPCM method, which is a lossless encoding that has a small compression ratio but can completely reproduce the original image.

[0014]

As an ideal encoding method, a method that has high encoding efficiency (without loss) without losing information of the original picture is desired. However, as described above, in the current DPCM method, There is a problem that the coding efficiency is not so high.

The present invention has been made in view of the above circumstances, and it is an object of the present invention to provide a multi-valued image encoding device capable of encoding a multi-valued image losslessly and efficiently.

[0016]

According to a first aspect of the present invention, there is provided a multi-valued image encoding apparatus, wherein each pixel constituting a multi-valued image is set as a target pixel, and a predicted value of the target pixel is set as a surrounding pixel. Pixel level prediction means for calculating from the pixel level, prediction error calculation means for calculating a prediction error value by subtracting the prediction value of the pixel of interest calculated by the pixel level prediction means from the actual pixel value of interest, and the prediction error Prediction error statistical means for obtaining statistical information by calculating the bias or distribution range of the prediction error value calculated by the calculation means, and a statistical model required for arithmetic coding based on the information statistically obtained by the prediction error statistical means Model creation means that adaptively changes the prediction error value using the learning effect, the correlation between each color component, the statistical information of the prediction error value, and the like, and the prediction error value calculated by the prediction error calculation means and the statistical model creation. Characterized by comprising an arithmetic coding means performing arithmetic coding using a statistical model created by means.

According to a second aspect of the present invention, there is provided a multi-valued image encoding apparatus, comprising: a color component separating unit for separating a color image composed of a plurality of color components into a multi-valued image for each of the color components; For each of the multi-valued images for each component, each pixel constituting the multi-valued image is set as a target pixel, and a prediction value of the target pixel is calculated from a surrounding pixel level, and the pixel level prediction unit calculates the prediction value of the target pixel. Prediction error calculating means for calculating a prediction error value by subtracting the predicted value of the target pixel thus calculated from the actual pixel value of interest, and obtaining a bias or distribution range of the prediction error value calculated by the prediction error calculating means. A prediction error statistical means for obtaining statistical information, and a statistical model required for arithmetic coding based on the information statistically obtained by the prediction error statistical means. A learning effect, a correlation between respective color components, and a statistical information of a prediction error value. A statistical model creating means that adaptively changes and creates the statistical code, and an arithmetic code that performs arithmetic coding using the prediction error value calculated by the prediction error calculating means and the statistical model created by the statistical model creating means And code generating means for generating the entire code string based on the code for each color component generated by the arithmetic coding means corresponding to the multi-valued image of each color component. I do.

According to a third aspect of the present invention, in the multi-level image encoding apparatus according to the second aspect, each of the statistical model creating means corresponding to each of the multi-level images of the respective color components has already been encoded. A statistical model is created based on information statistically obtained by the prediction error statistical means corresponding to the multi-valued image of the color component for which the above has been completed.

According to a fourth aspect of the present invention, in the multi-level image encoding apparatus according to the second aspect, the prediction error calculating means respectively corresponding to the multi-level image of each color component includes:
It is characterized in that a prediction error value is calculated in consideration of a correlation between color components.

According to a fifth aspect of the present invention, in the multi-valued image encoding apparatus according to the second aspect, the statistical model creating means corresponding to each of the multi-valued images of the respective color components comprises: A statistical model is created in consideration of the correlation of

According to a sixth aspect of the present invention, in the multi-level image encoding apparatus according to any one of the first and second aspects, the statistical model creating means includes a step of: Based on statistical information obtained by statistically calculating the prediction error value calculated by the error calculation unit, it is determined whether the target pixel to be coded is a pixel in a character area or a pixel in a photographic area. The reference pixel arrangement used in the statistical model is changed based on this.

According to a seventh aspect of the present invention, in the multi-valued image encoding apparatus according to any one of the first and second aspects, the statistical model creating means includes the prediction error statistical means. Based on statistical information obtained by statistically calculating the prediction error value calculated by the prediction error calculation unit, it is determined whether the target pixel to be coded is a pixel in a character area or a pixel in a photographic area. The number of reference pixels used in the statistical model is changed based on the result.

According to a eighth aspect of the present invention, in the multi-level image encoding apparatus according to any one of the first and second aspects, the statistical model creating means includes a step of: Based on statistical information obtained by statistically calculating the prediction error value calculated by the error calculation unit, it is determined whether the target pixel to be coded is a pixel in a character area or a pixel in a photographic area. The method is characterized in that the state of the prediction error level used in the statistical model is changed based on the statistical model.

[0024]

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

Before describing the multi-level image coding apparatus according to the present invention in detail, a facsimile apparatus will be described as an example of a case where the multi-level image coding apparatus according to the present invention is specifically applied. This will be described with reference to FIG. The figure shows
FIG. 1 is a block diagram of a facsimile apparatus on a transmission side and a reception side to which a multi-level image encoding device according to the present invention is applied.

First, on the transmission side, an image reading unit 101 reads a document using a CCD image sensor or the like, and a subsequent image processing unit 102 performs processing for converting transmission data into appropriate data. The encoded data created by encoding is transmitted to the transmission path. On the other hand, when an image is reproduced on the receiving side, the encoded data from the transmission path is decoded by the decoding unit 104, the image processing unit 105 performs image processing suitable for the output device, and the image output unit such as a plotter is used. A hard copy is obtained by outputting to 106.

Examples of the processing performed by the image processing units 102 and 105 include resolution conversion for a binary image,
There is a size conversion and the like, and in a multi-valued image including a color, a color (color component) conversion, a resolution conversion, a size conversion and the like can be mentioned. In addition, the above-described encoding unit 103 and decoding unit 104
As a conventional example of the encoding system used in the above, there are MH, MR, MMR system and JBIG system used in a conventional facsimile machine in the case of a binary image, and in the case of a multi-valued image, There is a JPEG system.

Among these encoding methods, the JPEG / Spatial method has already been standardized as an encoding method related to the multi-level image encoding device according to the present invention, and is capable of encoding a multi-tone image losslessly. Will be described briefly. The basics of JPEG are DCT (Discr.
This is a DCT method using an ET Cosine Transform. However, since the DCT method is a lossy coding method, a DPCM (Differential PCM) is used in a two-dimensional space as a method for implementing lossless coding.
A JPEG / Spatial method for performing the following is provided.

FIG. 2 shows a basic block diagram of the JPEG / Spatial system. An input image whose bit precision can take an arbitrary number of bits from 2 bits to 16 bits is input to a predictor 201, and a prediction error signal obtained by the predictor 201 is entropy-encoded and compressed by an entropy encoder 202. The encoded data is output.

FIG. 3 shows the prediction calculation formula in the predictor 201 and the positional relationship between the target pixel y to be coded at present and the three surrounding pixels (a, b, c) used for prediction. The prediction operation formula is as follows: y = a, y = b, y = c, y = a + b−
c, y = a + ((bc) / 2), y = b + ((a−
c) / 2) and y = (a + b) / 2, and a prediction error value is obtained by subtracting the prediction value calculated by each of the above prediction calculation expressions from the target pixel y.

The prediction error value thus calculated by the predictor 201 is entropy-coded by the entropy coder 202. The entropy coding includes:
There are Huffman coding and arithmetic coding.

In the case of Huffman coding, the prediction error value is
First, they are grouped according to the table shown in FIG.
By this grouping, a group number (SSSS) and an additional bit indicating the prediction error value within the group (the number of bits is S
SSS). And SS
The SS is encoded using a one-dimensional Huffman code table, and an additional bit is added after each Huffman code.

On the other hand, in the case of arithmetic coding, after a prediction error value is binarized, arithmetic coding is performed using a statistical model. First, the binarization will be described below. (1) Judgment of zero Outputs 0 when the prediction error is zero, and outputs 1 when the prediction error is not zero. (2) Sign (sign) judgment Outputs 0 when the prediction error is positive, and outputs 1 when the prediction error is not positive. (3) Group number Sz is defined as a value obtained by subtracting 1 from the absolute value of the prediction error, and it is determined whether or not the Sz belongs to group n (n = 0 to 15). This determination is performed in ascending order of the group number until the group to which Sz belongs is found. At this time, if Sz does not belong to group n, 1 is output, and if Sz belongs to group n, n is output. (See FIG. 5) (4) Additional bits The additional bits used for identifying the coefficients in the group are output as they are.

When the binary-coded expression sequence is arithmetically encoded by the above-described procedure, each determination item is classified and encoded as shown in FIG. 6 in order to estimate the probability of binary data. In the figure, the reason why there is 5 × 5 is that the prediction error of the pixel immediately above and the pixel immediately to the left is divided into five types according to the magnitude and the sign.

As shown in FIG. 7, in the arithmetic coding method, pixel information to be coded among input image data is input to an arithmetic coding circuit 402, and pixel information around the pixel to be coded is , Is input to a prediction information creation circuit (sometimes called a template) 401,
1 creates prediction information obtained by performing Markov separation of an information source according to the situation of a pixel to be coded and surrounding pixels, and dynamically evaluates prediction data in an arithmetic coding circuit 402 based on the prediction information. While encoding.

The arithmetic coding method performed by the arithmetic coding circuit 402 generally has better coding efficiency than the conventional run-length coding method (MH method, MR method, etc.). The encoding method is to make the corresponding sections ([0.0 ... 0, 0.1 ... 1] in binary decimal number) on the number line of (0, 1) unequal length according to the occurrence probability of each symbol. The coordinates of the points included in the section obtained by dividing and assigning the target symbol sequence to the corresponding partial section and repeating the division recursively can be at least distinguished from other sections.
It is expressed as a decimal number and used as a code as it is.

The concept of arithmetic coding will be briefly described with reference to FIG. 8 using the symbol sequence "0100" as an example. In the figure, first, when encoding the first symbol, the entire section is divided into A (0) and A (1) according to the ratio of the occurrence probabilities of the symbols 0 and 1, and the section A (0) is selected by the occurrence of 0. Is done.
Next, when encoding the second symbol, A (0) is further divided by the occurrence probability ratio of both symbols in that state, and A (0) is defined as a section corresponding to the generated symbol sequence.
1) is selected. Encoding proceeds by repeating such division and selection processes.

The coded data thus coded is
The image data input to the arithmetic decoding circuit 403 and decoded is output as decoded pixel information, and is input to the prediction information creating circuit 404 for decoding the next coded data. 403 performs decoding while evaluating the encoded data based on the prediction information from the prediction information creation circuit 404.

In the decoding in that case, a process completely opposite to the encoding is performed, and the symbol is reproduced based on the binary decimal number indicated by the code. What is important at this time is the width of the number line when encoding the symbol, and if the width of the number line does not match between the start of encoding and the start of decoding, the symbol cannot be accurately reproduced. Usually, the width of this number line is set to 1 on the encoding side and the decoding side.

Next, a multi-level image encoding apparatus according to an embodiment of the present invention will be described with reference to the block diagram of FIG.

In the figure, the multi-level image encoding apparatus sets each pixel constituting the input multi-level image information as a target pixel, and calculates a predicted value of a pixel level of the target pixel from a peripheral pixel level. The prediction means 601, the prediction error calculation means 602 for calculating a prediction error value by subtracting the predicted value of the pixel of interest calculated by the pixel level prediction means from the actual pixel value of interest, and the prediction error calculation means 602 Prediction error statistical means 60 for statistic of prediction error value
3. Statistical model creation means 604 for adaptively creating a statistical model required for arithmetic coding based on the information statistically obtained by the prediction error statistical means 604, and the prediction calculated by the prediction error calculation means 602. An arithmetic coding unit 605 that performs arithmetic coding using the error value and the statistical model created by the statistical model creating unit 604 and outputs code data.

The operation of the multi-level image coding apparatus thus configured will be briefly described. First, the image information input for encoding is input to the pixel level prediction means 601 and the prediction error calculation means 602. The pixel level prediction means 601 predicts the pixel level of the target pixel from the pixel level values of surrounding pixels. As an example of the prediction method, if the prediction level value is Px, the level value of the pixel immediately to the left of the target pixel is Ra, and the pixel level value of the pixel immediately above is Rb, Px = (Ra +
There is a method of calculating Px by the formula of Rb) / 2.

The prediction error calculating means 602 calculates a prediction error value from the predicted prediction level value and the actual pixel level value of the target pixel. Assuming that the prediction error value is Dx and the target pixel level value is P, the calculation method of the prediction error value is simply Dk
= Px-P. The calculated prediction error value is arithmetically encoded by the arithmetic encoding unit 605, and before that, a statistical model is created.

The prediction error statistic means 603 obtains statistical information by using the prediction error value as reference information for creating a statistical model and performing statistics. As an example of the statistical information, it is possible to statistically predict the bias of the prediction error value and the distribution range of the prediction error value. The statistical information is input to the statistical model creating means 604 to create a statistical model for arithmetic coding. Note that the created statistical model changes adaptively based on the statistical information. The method of adaptively changing the value will be described later. The arithmetic coding unit 605 performs arithmetic coding using the statistical model created by the statistical model creating unit 604 and outputs code data. A code string for the input image information is created as described above.

Next, another multi-level image encoding apparatus according to the embodiment of the present invention will be described with reference to the block diagram of FIG. The multi-level image coding apparatus shown in FIG. 9 is directed to a simple multi-level image in FIG. 9, whereas R (red), G (green), and B (blue) Although the target is color multi-valued image information composed of primary color components, the configuration and type of color components may be other.

Referring to FIG. 10, the multi-valued image coding apparatus includes a color component separating unit 701 for separating input color multi-valued image information into a multi-valued image for each color component. Pixel level predicting means (702a, 702b, 702c) for calculating a predicted value of a pixel level of a pixel of interest from surrounding pixel levels for each of the value plane information, and pixel level prediction for each of multi-valued image information decomposed for each color component Prediction error calculating means (703a, 703b, 703c) for calculating a prediction error value by subtracting the prediction level value of the pixel of interest calculated by the means (702a, 702b, 702c) from the actual pixel level value of interest, for each color component For each of the multi-valued image information decomposed into two, the prediction error value calculated by the prediction error calculation means (703a, 703b, 703c) is statistically calculated.認差 flow meter means (704a, 704
b, 704c), prediction error statistical means (704a, 7) for each of the camera device surface information separated for each color component.
A statistical model creating means (705a, 705b, 705c) for adaptively changing and creating a statistical model required for arithmetic coding based on the information statistically obtained in steps 04b and 704c);
The prediction error value calculated by the prediction error calculation means (703a, 703b, 703c) for each of the multi-valued image information decomposed for each color component and the statistical model creation means (705a,
Arithmetic coding means (706a, 706) for performing arithmetic coding using the statistical model created in 705b, 705c)
b, 706c), their arithmetic coding means (706a,
It comprises a code generation means 707 for generating an entire code string for the input color multi-valued image based on the code for each color component multi-valued image generated in 706b, 706c).

The operation of the multi-level image coding apparatus thus configured will be briefly described. First, the color multi-valued image information input for encoding is input to the color component decomposing unit 701 and decomposed into a multi-valued image for each color component. The decomposed multi-valued image for each color component is processed for each color component.
The surface information for each of the separated color components is supplied to the pixel level prediction means (7
02a, 702b, 702c) and prediction error calculation means (703a, 703b, 703c). The pixel level prediction means (702a, 702b, 702c) predicts the pixel level of the target pixel from the pixel level values of surrounding pixels. As an example of the prediction method, as in the multi-level image encoding device illustrated in FIG. 9, when the prediction level value is Px, the level value of the pixel immediately to the left of the target pixel is Ra, and the pixel level value of the pixel immediately above is Rb, Px = (Ra + Rb) / 2
There is a method for obtaining x. Prediction error calculation means (70
In 3a, 703b, and 703c), a prediction error value is calculated from the predicted prediction level value and the actual pixel level value of the target pixel. As an example of a method of calculating the prediction error value, the prediction error value is calculated as D
x, and the target pixel level value is P, simply Dk = Px
-P. The calculated prediction error value is arithmetically encoded by the arithmetic encoding means (706a, 706b, 706c). Before that, a statistical model is created.

The prediction error statistical means (704a, 704b,
In step 704c), the prediction error value is used as reference information for creating a statistical model, the statistics are obtained, and statistical information is obtained.
As an example of the statistical information, there is a method of statistically calculating the bias of the prediction error value and the distribution range of the prediction error value. The statistical information is stored in statistical model creation means (705a, 705b, 705).
A statistical model for arithmetic coding is input to c) and is created. Note that the created statistical model changes adaptively based on the statistical information. The method of adaptively changing the value will be described later. Arithmetic coding means (706a, 706b, 70
6c) is a statistical model creation means (705a, 705b,
The arithmetic coding is performed using the statistical model created in 705c), and code data is output. As described above, the code sequence for the multi-valued image of each color component is created. Finally, each arithmetic coding means (706a, 706b, 706c)
The code string for the multi-valued image for each color component created in step (1) is input to the code creation means 707, and the entire code string is created.

In the multi-valued image coding apparatus having the configuration shown in FIG. 10, in order to speed up the processing time, a bus of the coding means is provided in parallel for each multi-valued image of each of the RGB color components. Although they are arranged, as shown in FIG. 11, it is also possible to recursively use the bus of one encoding means for the multi-valued image of each color component. The advantages of the configuration shown in FIG. 11 are that the configuration is simplified and that the learning effect described later can be fully utilized. However, the learning effect can be used by exchanging information between the statistical model creating means (705a, 705b, 705c) even in the configuration shown in FIG.

The learning effect will be described. When the buses of the encoding means are simply arranged in parallel for each multi-valued image of each color component of RGB as in the multi-valued image encoding device having the configuration shown in the tenth, the statistical model is independent for each color component. However, since the learning effect of the statistical model on the color components that have already been encoded cannot be used, there is a problem that the encoding efficiency is still low. In order to change the statistical model while performing encoding, it is necessary to dynamically judge from the state of the prediction error value. In other words, learning can create an optimal model.
Therefore, in order to effectively use the learning in the already completed color component, by creating a statistical model in the encoding of the subsequent color component based on the information of the statistical model in the already completed color component, Lossless and highly efficient encoding of color images is possible.

Next, in the multi-level image coding apparatus shown in FIG. 9, 10 or 11, the statistical model creating means uses the statistical model required by arithmetic coding based on the information statistically obtained by the prediction error statistical means. To create an adaptively modified
A method for adaptively changing the statistical model will be described below. Note that the reference numerals given to the respective components shown in FIG. 9, 10 or 11 are basically omitted, and are added only when necessary.

As a method of adaptively changing the statistical model, in addition to using the learning effect described above, a case where the correlation between the respective color components is used as a reference condition for adaptively changing the statistical model. (Limited to the multi-level image encoding apparatus shown in FIG. 10 or 11) and the case of using the statistical information of the prediction error value (for all of the multi-level image encoding apparatuses shown in FIG. 9, 10 or 11). Applicable).

First, the case where the correlation between the respective color components is used will be described. There is some correlation between the color components constituting the color image. Therefore, the statistical model is not fixed as in the conventional statistical model shown in FIG. 6, but is created by changing the statistical model more suitable for encoding by considering the correlation between the color components. , The coding efficiency can be improved. In this case, in the multi-level image encoding device shown in FIG. 10, the statistical model creation means (705a, 705b, 705c) corresponding to each color component is used as the prediction error statistical means (704a, 704b,
The correlation of each color is obtained by considering not only 704c) but also the statistical information from the prediction error statistical means (704a, 704b, 704c) corresponding to other colors, and a statistical model is created in consideration of the correlation. In the multi-level image encoding device shown in FIG. 11, the statistical model creation unit 805 obtains the correlation of each color by considering the statistical information for each color component from the prediction error statistical unit 804, and obtains a statistical model in which the correlation is considered. Create

Next, a case where the statistical information of the prediction error value is used will be described. As a method of changing the statistical model according to the statistical information of the prediction error value, it is determined whether the pixel of interest is in the character region or the photograph region from the statistical information of the prediction measurement error value, and the method is suitable for the character region or the photograph region. There is a method to change to a statistical model.

That is, as for the prediction error value Pd in the pixel of the character area, Pd = 0 in the character portion and the background portion, and Pd in the edge portion, in many cases, are relatively large values. As for the distribution of Pd, Pd = 0 and Pd concentrate on relatively large values. On the other hand, Pd in the photographic area is opposite to that in the character area, and Pd = 0 is relatively small. On the other hand, the value of Pd is rarely increased, and the distribution is concentrated on a small value of Pd. Therefore, if this property is used, it is possible to create an optimal statistical model for the target pixel, and it is possible to improve coding efficiency.

There are three factors for creating a statistical model: the arrangement and number of reference pixels and the classification of prediction error levels. That is, these factors may be adaptively changed in order to adaptively change and create the statistical model. Regarding the arrangement of reference pixels, when encoding a character image, it is better to arrange the reference pixels relatively close to the pixel of interest to improve the encoding efficiency. In some cases, the encoding efficiency may be improved when the arrangement is located at a position distant from the pixel of interest. Therefore, this property is used. It is not necessary to increase the number of reference pixels when encoding a character image, and if the number of states is increased, the encoding efficiency is rather deteriorated. Conversely, in a photographic image, a relatively large number may improve the coding efficiency in some cases. Since the classification of the prediction error level differs depending on the image, the classification suitable for the image is performed.

Although the description has been made mainly on adaptively changing the statistical model in order to improve the coding efficiency in the arithmetic coding means, the prediction level value of the target pixel by the pixel level prediction means and the actual It is also important to make the difference from the pixel value of the target pixel as small as possible in order to improve the coding efficiency. Therefore, in the multi-level image coding apparatus having the configuration shown in FIG. 10 or 11, the prediction level of the target pixel is set to a more accurate value by considering the correlation between the color components forming the color multi-level image. It is also possible to improve the coding efficiency by reducing the prediction error value as a whole. In that case, in the multi-level image encoding device shown in FIG. 10, the pixel level prediction means (702a, 702b,
Each of 702c) notifies the surrounding pixel level of each color component, thereby obtaining the correlation of each color, and calculating the prediction error value in consideration of the correlation. In the multi-level image encoding device shown in FIG. 11, the pixel level prediction means 802 obtains the correlation of each color by comparing the attention pixel level of each color component,
A prediction error value is calculated in consideration of the correlation.

[0058]

According to the first aspect of the present invention, the lossless and highly efficient encoding of a monochrome multi-tone image is performed by adaptively changing the statistical model from the distribution range and deviation of the encoding target information. It becomes possible. In other words, in arithmetic coding, if an appropriate statistical model is used, very high coding efficiency can be obtained. On the other hand, since a statistical model is very related to the type and properties of an image, By efficiently changing the statistical model by using the distribution range and deviation of the encoding target information as a judgment material in consideration of the fact that the encoding process is different, efficient encoding becomes possible.

According to the second aspect of the present invention, the statistical model is adaptively changed for each color component based on the distribution range and deviation of the information to be encoded, so that lossless and highly efficient encoding is performed in a color multi-valued image. Becomes possible. In other words, by encoding a color multi-valued image into color components, each color component can be treated in the same way as a black-and-white multi-tone image. It is possible to perform encoding with very high encoding efficiency.

According to the third aspect of the present invention, in order to effectively use the learning on the already-encoded color components, the following is performed based on the information of the statistical model on the already-encoded color components. By creating a statistical model for encoding color components, lossless and highly efficient encoding of a color image can be performed. In other words, in the invention according to claim 2, since the statistical model is created independently for each color component, the learning effect of the statistical model on the color components that have already been encoded cannot be used, and the encoding efficiency cannot be improved. Although there is room for improvement, in the present invention, the coding efficiency can be further improved by using the learning effect of the statistical model on the color components that have already been coded.

According to the fourth aspect of the invention, by calculating the prediction error value using the correlation between the color components, it is possible to perform lossless and highly efficient encoding on a color image. In other words, in the invention according to claim 2, there is still room for improvement of the coding efficiency because the correlation between the color components is not used. By paying attention to the existence of the correlation and considering the correlation between the color components, the prediction level of the target pixel can be set to a more accurate value. Since the prediction error value becomes smaller as the prediction level becomes more accurate, the coding efficiency is improved as a result.

According to the fifth aspect of the present invention, by creating a statistical model using the correlation between color components, lossless and highly efficient coding of a color image can be performed. In other words, the present invention focuses on the existence of a correlation between the color components forming a color image, and creates a statistical model in consideration of the correlation between the color components, thereby applying a more optimal statistical model. Arithmetic coding can be performed, and as a result, coding efficiency is improved.

According to the invention of claim 6, it is determined whether the pixel is a character area pixel or a photograph area pixel from the statistical information of the prediction error value, and the reference pixel arrangement is changed based on the determination result. By adaptively changing the statistical model, it is possible to perform lossless and highly efficient coding. That is,
In the present invention, in the statistical model in arithmetic coding, when encoding a character image, it is better to place the reference pixels relatively close to the pixel of interest to improve encoding efficiency, while in a photographic image, Conversely, considering that the placement of the reference pixels may be better placed at a position distant from the pixel of interest, the statistical efficiency of the reference pixel arrangement is applied to the character image, A more efficient coding can be performed on a photographic image by applying a statistical model of a reference pixel arrangement suitable for the photographic image.

According to the seventh aspect of the present invention, it is determined from the statistical information of the prediction error value whether the pixel is a character area pixel or a photograph area pixel, and the number of reference pixels is changed based on the determination result. By adaptively changing the statistical model in
Lossless and highly efficient encoding can be performed. That is, in the present invention, in the statistical model in arithmetic coding, when encoding a character image, the number of states does not need to be very large, and if the number of states is increased, the coding efficiency is rather deteriorated. On the other hand, taking into account that encoding efficiency is better when the number of states is relatively large in a photographic image, a statistical model of the number of reference pixels suitable for the character image is applied to a character image,
Applying a statistical model of the number of reference pixels suitable for the photographic image enables more efficient encoding.

According to the invention of claim 8, it is determined whether the pixel is a character area pixel or a photograph area pixel from the statistical information of the prediction error value, and the state of the prediction error level is classified based on the determination result. By changing the statistical model adaptively by the change, lossless and highly efficient coding can be performed. That is, in the present invention, in the statistical model in arithmetic coding, the arrangement and number of reference pixels and the classification of the prediction error level are important, and the classification of the prediction error level differs depending on the image. Considering that it is possible to increase the coding efficiency by performing the classification, a statistical model in which the state of the prediction error level suitable for that is applied to the character image, and to the photographic image, By applying a statistical model in which a state of an appropriate prediction error level is classified, more efficient coding can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram of a facsimile apparatus to which a multi-level image coding apparatus according to the present invention is applied.

FIG. 2 shows a basic block diagram of the JPEG / Spatial system.

FIG. 3 shows a pixel of interest y and three surrounding pixels (a,
It is a figure which shows the positional relationship with b, c).

FIG. 4 is a diagram showing grouping of prediction error values.

FIG. 5 is a diagram showing grouping of Sz.

FIG. 6 is a diagram illustrating a lossless encoded statistical model by arithmetic encoding.

FIG. 7 is a block diagram illustrating an example of an arithmetic coding method.

FIG. 8 is a conceptual diagram of arithmetic coding.

FIG. 9 is a block diagram of a multi-level image encoding device according to an embodiment of the present invention.

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

FIG. 11 is a block diagram of another multi-level image encoding device according to an embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 101 Image reading part 102 Image processing part 103 Encoding part 104 Decoding part 105 Image processing part 106 Image output part 201 Predictor 202 Entropy coder 401 Prediction information preparation circuit 402 Arithmetic coding circuit 403 Arithmetic decoding circuit 404 Prediction information preparation circuit 601, 702a, 702b, 702c, 802 Pixel level prediction means 602, 703a, 703b, 703c, 803 Prediction error calculation means 603, 704a, 704b, 704c, 804 Prediction error statistical means 604, 705a, 705b, 705c, 805 Statistical model Creation means 605, 706a, 706b, 706c, 806 Arithmetic coding means 701, 801 Color component separation means 707, 807 Code creation means

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H04N 1/41-1/419

Claims (8)

(57) [Claims] 1. A pixel level predicting means for calculating a predicted value of a pixel of interest from surrounding pixel levels using each pixel constituting a multi-valued image as a pixel of interest; A prediction error calculating means for calculating a prediction error value by subtracting the prediction value from the actual pixel value of interest, and a prediction for obtaining statistical information by obtaining a bias or a distribution range of the prediction error value calculated by the prediction error calculating means. Error statistical means and the statistical model required for arithmetic coding based on the information statistically obtained by the predictive error statistical means, adaptively utilizing learning effects, correlation between color components, statistical information of predictive error values, etc. A statistical model creating means for creating by changing to the above, an arithmetic code for performing arithmetic coding using the prediction error value calculated by the prediction error calculating means and the statistical model created by the statistical model creating means Multi-level image coding apparatus characterized by comprising a means. 2. A color component decomposing means for decomposing a color image composed of a plurality of color components into a multi-value image for each of the color components, and for each of the decomposed multi-value images for each of the color components, Each pixel constituting the multi-valued image is set as a target pixel,
A pixel level prediction unit for calculating the predicted value of the target pixel from the surrounding pixel level; and a prediction error value is calculated by subtracting the predicted value of the target pixel calculated by the pixel level prediction unit from the actual target pixel value. A prediction error calculating means, a prediction error statistical means for obtaining statistical information by obtaining a bias or a distribution range of the prediction error value calculated by the prediction error calculating means, and Statistical model creation means for creating a statistical model required for arithmetic coding by adaptively changing the learning effect, correlation between respective color components, statistical information of prediction error values and the like, and the prediction error calculation means An arithmetic coding unit for performing arithmetic coding using the calculated prediction error value and the statistical model created by the statistical model creating unit, and a multi-valued image of each of the color components. A multi-valued image encoding apparatus, comprising: code creating means for creating an entire code sequence based on codes for respective color components created by corresponding arithmetic coding means. 3. The statistical model creating means respectively corresponding to the multi-valued image of each color component, based on the information statistically obtained by the prediction error statistical means corresponding to the multi-valued image of the color component which has already been encoded. 3. The multi-valued image encoding device according to claim 2, wherein the statistical model is created by performing the calculation. 4. The multi-valued image according to claim 2, wherein said prediction error calculating means respectively corresponding to the multi-valued image of each color component calculates a prediction error value in consideration of a correlation between the color components. Encoding device. 5. The multi-valued image according to claim 2, wherein said statistical model generating means corresponding to each of the multi-valued images of each color component generates a statistical model in consideration of a correlation between the color components. Encoding device. 6. The statistical model creating means according to claim 1, wherein said target pixel to be coded is a character area based on statistical information obtained by said predictive error statistical means statistically calculating the predictive error value calculated by said predictive error calculating means. 2. The method according to claim 1, further comprising: determining whether the pixel is a pixel in the photographic area or not, and changing a reference pixel arrangement used in the statistical model based on the determination result.
Or the multi-level image encoding device according to any one of 2. 7. The statistical model creating means according to claim 1, wherein said pixel of interest to be encoded is a character area based on statistical information obtained by said prediction error statistical means statistically calculating the prediction error value calculated by said prediction error calculating means. 3. The method according to claim 1, wherein it is determined whether the pixel is a pixel of the photographic area or the pixel of the photographic area, and the number of reference pixels used in the statistical model is changed based on a result of the determination. Multi-level image encoding device. 8. The statistical model creating means according to claim 1, wherein said target pixel to be coded is a character area based on statistical information obtained by said predictive error statistical means statistically calculating the predictive error value calculated by said predictive error calculating means. 3. The method according to claim 1, wherein a determination is made as to whether the pixel is a pixel of a photographic area or a pixel of a photographic area, and the state of the prediction error level used in the statistical model is changed based on the determination result. Multi-level image encoding device.