JP2007295130A - Image data encoder, program, computer-readable recording medium, and image data encoding method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To achieve an image data encoder encoding image data including a plurality of objects having different image types at a high compression rate. <P>SOLUTION: The image data encoder 1 imitates the QM-Coder of JBIG. In this case, the image data encoder 1 comprises: a region determination section 22 that detects the image type of a region including a pixel to be encoded extracted by a context generator 21 and determines whether the detected result differs from a previous one; and a table change section 23 for resetting a learning table 32 stored in a recorder 31 when the region determination section 22 determines that the detected result differs from the previous one. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an image data encoding apparatus that encodes image data by arithmetic encoding.

  As a method of modeling binary image data with a Markov information source, predicting an encoding target symbol based on the state of its surrounding symbols, and performing arithmetic encoding according to the prediction result, JBIG (ITU-T T.82) QM-Coder is known (see Non-Patent Documents 1 and 2).

  This QM-Coder converts a binary input signal (input bit) sequence constituting image data into a real number x of 0 or more and less than 1. The QM-Coder changes the range (hereinafter referred to as “interval”) in which the real number x as the encoding result can exist according to a predetermined procedure in accordance with the input bits. If the interval does not change despite the input of a new bit, the low number that is the minimum value of the interval is used as the encoding result.

  More specifically, the QM-Coder calculates the predicted value (dominance probability symbol (MPS)) of the next (current) input bit from the input bit string up to the previous time. Further, according to this prediction, the interval is divided into two regions according to the ratio between the appearance probability of the dominant probability symbol and the appearance probability of the inferior probability probability symbol (LPS). Here, of the two divided regions, the region closer to 0 is the MPS interval corresponding to the appearance probability of the dominant probability symbol, and the region far from 0 is the LPS interval corresponding to the appearance probability of LPS. And

  The QM-Coder changes the interval so as to remove the LPS interval from the current interval when the input bit matches the MPS. On the other hand, if the input bit and the MPS do not match, the interval is changed so as to remove the MPS interval from the current interval. A low number of intervals when processing for all input strings is completed is a real number x, and the real number x is an encoding result. The above procedure does not change the low number of intervals when the input bits match the expected input bits. For this reason, there is no change in the output (encoding result), and there is no need to output a new bit, so encoding is performed using this.

  When performing the above-described procedure, if the encoding process is performed, the interval size gradually decreases, and eventually cannot be expressed with the fixed accuracy of the computer. Therefore, a process called normalization processing is performed in which the size of the interval is doubled until the size of the interval becomes 0.5 or more.

Here, in the QM-Coder, the size of the interval is held in the register A (Areg) in the memory area, and the low number of intervals is held in the register C (Creg) in the memory area. In addition, the interval from 0 to 1 is multiplied by 2 ¹⁶ (= 65536) at the time of implementation, that is, 0x0 (= 0; “0x” represents a hexadecimal number) to 0x10000 (= 65536). It is expressed by 16 bits. In this case, the register A contains a normalized interval so that the size is 0.5 (0x8000) or more and 1 or less. Each time the value stored in the register A is doubled by the normalization process, the value stored in the register C is also doubled. For this reason, the influence on the output caused by doubling the value of the register A is offset.

In order to calculate the LPS interval,
(Interval size) x (LPS appearance probability)
However, in the QM-Corder implementation, the value of this LPS interval is approximated by an estimated value (LSZ) obtained statistically. This LSZ is given as a table associated with the previous input bit string used for prediction.

  In summary, in the implementation form of QM-Coder, when MPS is given as an input, LSZ is subtracted from the interval, and when LPS is given as an input, the interval is set to LSZ. Also, normalization is performed when the interval of this operation becomes 0.5 or less. When normalization is performed, the value of LSZ is updated. Moreover, the LPS interval may become larger than the MPS interval due to the above-described operation. At this time, the MPS interval and the LPS interval are exchanged.

  FIG. 12 is a block diagram showing a functional configuration of a conventional encoding apparatus in which JBIG QM-Coder is implemented. FIG. 13 is a flowchart showing the flow of processing by the conventional encoding apparatus. As shown in FIG. 12, the conventional encoding device 100 generally includes a context generation unit 122, an acquisition unit 123, an arithmetic encoding unit 129, a recording unit 125, and a register 126.

  When binary image data is input to the encoding apparatus 100, the context generation unit 122 divides the input image into at least one stripe (S150). The stripe is a unit for performing the encoding process, and is composed of one or more horizontal lines in the image. When the processing for all stripes is completed by the loop for each stripe (LOOP_A) from S150 to S164, the image encoding processing is completed.

  The context generation unit 122 selects a pixel to be encoded (encoding target pixel) and reads the value (symbol) (S151). Here, the target pixel is selected sequentially from the upper left to the lower right of the stripe. When the encoding of the target pixel is performed by the loop (LOOP_B) of S151 to S162 and the encoding process is completed for all the pixels included in the stripe, the process of one stripe is completed. Subsequently, the context generation unit 122 reads the values of a plurality of peripheral pixels existing around the target pixel, and extracts the context (S152). The context indicates the pattern of the peripheral pixel group.

  Next, the acquisition unit 123 accesses the learning table 127 of the recording unit 125 to acquire the index ST of the appearance probability estimation value of the MPS and the inferior probability symbol corresponding to the context extracted in S152 (S153).

  Next, the preprocessing unit 129d determines whether or not the value of the target pixel is MPS based on the value (symbol) of the target pixel read in S151 and the MPS acquired in S153 (S154). When it is determined in S154 that the value of the target pixel is MPS, the MPS processing unit 129a performs MPS processing (CODE_MPS: S155 to S158) in the arithmetic coding unit 129. On the other hand, when it is determined in S154 that the value of the target pixel is LPS, the LPS processing unit 129b included in the arithmetic coding unit 129 performs LPS processing (CODE_LPS: S159 to S161).

  The MPS processing unit 129a accesses the probability estimation table 128 of the recording unit 125 to acquire the appearance probability estimated value (LSZ) of the inferior probability symbol corresponding to the index ST acquired in S153, and according to this, the register 126 stores The Areg 126a and the Creg 126b are rewritten (S155). Then, the MPS processing unit 129a determines whether or not the value of the Areg 126a is less than 0.5 (S56). If it is less than 0.5, the process proceeds to S157, and if it is 0.5 or more, The process proceeds to S162. In S157, the MPS processing unit 129a instructs the RENORM processing unit 129c, and the RENORM processing unit 129c normalizes the Areg 126a and the Creg 126b of the register 126, and outputs the obtained encoded data as necessary. Subsequently, the MPS processing unit 129a refers to the probability estimation table 128 of the recording unit 125, updates the learning table 128 of the recording unit 125 according to the determination result in S154 (S158), and proceeds to S162.

  On the other hand, in S159, the LPS processing unit 129b accesses the probability estimation table 128 of the recording unit 125 to acquire the probability estimation value of the inferior probability symbol corresponding to the index ST acquired in S153, and the register 126 accordingly. Areg 126a and Creg 126b are rewritten (S159). Then, in S160, the LPS processing unit 129b performs switching between LPS and MPS and code output for the Areg 126a and Creg 126b of the register 126 according to the value of SWITCH of the probability estimation table 128. That is, the LPS processing unit 129b determines whether or not to perform the replacement depending on whether or not the value of SWITCH in the probability estimation table 128 corresponding to the index ST is 1. The LPS processing unit 129b refers to the probability estimation table 128 of the recording unit 125, updates the learning table 128 of the recording unit 125 according to the determination result in S154 and the value of SWITCH (S161), and proceeds to S162.

  In S162, the arithmetic encoding unit 129 instructs the context generation unit 122 to increment the pixel, and the loop for the pixel ends. By performing such a procedure for all the pixels of the stripe, the encoding of the stripe is performed. In S163, it is determined whether or not the processing for the stripe has been completed. If it has been completed, the register 126 outputs the encoded data (flash processing: S163). Thus, before proceeding to the encoding of the next stripe, the portion of the encoding result so far that has not been output is output. In S164, the context generation unit 122 selects a new stripe. If encoding of all stripes is completed, it means that encoding of the entire image has been completed. If all of the input image data has been encoded, the process is terminated.

  As described above, in the QM-coder using the learning type arithmetic coding, the estimated value (LSZ) of the symbol appearance probability based on the context is determined based on the learning table 127 and the probability estimation table 128. The learning table 127 is updated based on the update contents shown in the probability estimation table 128 depending on whether the value of the encoding target pixel is MPS or LPS. Therefore, the learning table 127 is optimized as the input image data is encoded, and as a result, the compression rate of the image data is improved.

By the way, in Patent Document 1, when arithmetic coding is performed, tables indicating symbol appearance probability estimation values corresponding to contexts are used as images for document images, halftone dots or dither images, and error diffusion images. There has been proposed an image encoding device that prepares a plurality of types for each type and selects a table to be used for arithmetic encoding based on a result of region determination on an input image. Note that in the image encoding device described in Patent Document 1, region determination is performed by obtaining region information from input image data using a statistical method.
ITU-T (International Telecommunication Union Telecommunication standardization sector) WHITE BOOK Digital still image compression coding related recommendation (New Japan ITU Association) Recommendation T.82 WTSC-93 (WORLD TELECOMMUNICATION STANDARDIZATION CONFERENCE), Helsinki, 1-12 MARCH 1993 Japanese Patent Laid-Open No. 9-135358 (published on May 20, 1997)

  However, the above-described conventional QM-Coder has a problem that a satisfactory compression rate cannot be obtained when an image in which a plurality of different objects such as characters, graphics, and photographs are mixed is encoded.

  An image that is the basis of input image data often includes a plurality of types of partial images. For example, an image in which characters, figures, photographs, etc. are arranged in various places may be created. As described above, when a plurality of different types of objects are included in the image to be encoded, the learning rate of the learning table 127 tends to decrease in the boundary region between the objects. This problem will be specifically described as follows.

  For example, assume that pixels included in a character area are sequentially encoded in input image data. At this time, the learning table 127 is optimized. In this case, the learning table 127 is optimized so that the symbol appearance probability for the pixels in the character area can be estimated with high accuracy. Here, when the encoding target pixel shifts from the character region to the photo region, the appearance probability of the symbol in the pixel of the photo region cannot be accurately estimated by the learning table optimized for the character region. Therefore, it is necessary to optimize the learning table 127 again for the photographic area. Here, the compression ratio is lowered until the learning table 127 is optimized for the photographic area.

  Note that the learning table 127 optimized for the character area may be less accurate in estimating the appearance probability of the symbols in the pixels in the photographic area than the unlearned state, and may require a long period for optimization. Not a few. This is because an unlearned learning table is usually not specialized for each area but is set to average contents.

  For this reason, in the conventional QM-Coder, when encoding image data in which objects of different image types are mixed, the compression rate decreases in the boundary region between the objects, and a satisfactory compression rate is obtained. I can't get it.

  Patent Document 1 describes determining an image area including a pixel to be encoded. However, Patent Document 1 describes a statistical method based on input image data in order to determine an image area. Is adopted. However, in order to statistically determine an image region from input image data, a large amount of calculation and statistical data are required.

  As described above, the present invention has been made in view of the above problems, and its main purpose is to encode data with a high compression rate even for image data in which objects of a plurality of different image types are mixed. An object of the present invention is to realize an image data encoding device capable of performing the above.

  A further object of the present invention is to realize an image data encoding device that can easily and accurately detect what image type region in an image the data to be encoded is derived from. It is in.

  In order to solve the above-described problems, an image data encoding apparatus according to the present invention uses symbols constituting image data as encoding target symbols, and uses the encoding target symbols as symbol appearances estimated from surrounding symbol patterns. Learning that stores a learning table in which a correspondence relationship between a pattern of the peripheral symbols and an index for specifying the symbol appearance probability is stored in an image data encoding device that sequentially encodes by arithmetic encoding based on a probability Probability estimation storing a probability estimation table in which a correspondence relationship between table storage means, the index shown in the learning table, the symbol appearance probability, and update information indicating the update contents of the index in the learning table is shown Table storage means, and the encoding target symbol and the upper Extraction means for extracting a pattern of peripheral symbols, image type detection means for detecting an image type of a partial region in the image corresponding to the encoding target symbol extracted by the extraction means, and the image type detection means Learning table changing means for changing the learning table stored in the learning table storage means when the image type detected by the above is different from the image type previously detected by the image type detecting means; The index corresponding to the pattern of the peripheral symbols extracted by the extraction unit is acquired from the learning table, the appearance probability and the update information corresponding to the acquired index are acquired from the probability estimation table, and the acquired appearance probability The encoding target symbol extracted by the extracting means is calculated based on An encoding unit that performs encoding by encoding and updates the index included in the learning table based on the encoding target symbol and the acquired update information; the extraction unit; the image type detection unit; the learning Table changing means and loop means for causing the encoding means to repeatedly execute processing are provided.

  In order to solve the above problems, an image data encoding method according to the present invention uses symbols constituting image data as encoding target symbols, and the encoding target symbols are estimated from patterns of surrounding symbols. In an image data encoding method for sequentially encoding by arithmetic encoding based on a symbol appearance probability, an extraction unit extracts the encoding target symbol and the peripheral symbol pattern from the image data, and an image An image type detecting unit for detecting an image type of a partial region in the image corresponding to the encoding target symbol extracted in the extracting step; and the image type detected in the image type detecting step. Is different from the image type previously detected in the image type detection step A learning table changing step, wherein the learning table changing means changes the learning table stored in the storage means and indicating the correspondence relationship between the pattern of the surrounding symbols and the index for specifying the symbol appearance probability; The encoding means acquires the index corresponding to the pattern of the peripheral symbols extracted in the extraction step from the learning table stored in the storage means, and the index and the symbol shown in the learning table Acquire the appearance probability and the update information corresponding to the acquired index from the probability estimation table showing the correspondence between the appearance probability and the update information indicating the update contents of the index in the learning table, and the acquired appearance Extracted in the above extraction step based on probability An encoding step that encodes the encoding target symbol by arithmetic encoding and updates the index included in the learning table based on the encoding target symbol and the acquired update information, and the extraction step The image type detecting step, the learning table changing step, and the encoding step are repeatedly executed.

  According to the above configuration, the image type detection unit detects the image type (for example, text, vector graphics, bitmap, etc.) of the partial area in the image corresponding to the encoding target symbol. When the image type detected by the image type detection unit is different from the previously detected image type, the learning table stored in the learning table storage unit is changed by the learning table change unit. . Therefore, when the partial area in the image corresponding to the encoding target symbol shifts to an area of a different image type in the image, this can be detected by the image type detection means. At this time, since the learning table is changed, the learning table optimized for the old image type so far is used as it is, and the data of the new image type is not encoded as the encoding target symbol. Therefore, the period necessary for optimizing the learning table for a new image type is shortened, and as a result, it is possible to suppress a decrease in the compression rate in the boundary region between objects of different image types. Therefore, even image data in which objects of a plurality of different image types are mixed can be encoded at a high compression rate.

  The image data encoding device further includes an initial table storage unit that stores an initial table serving as a template for the learning table, and the learning table changing unit includes the image type detected by the image type detection unit. The learning table stored in the learning table storage means is stored in the initial table stored in the initial table storage means when the image type is different from the image type previously detected by the image type detection means. You may rewrite based on it.

  As described above, when a learning table optimized for a certain image type is used for encoding image data of another image type, the estimation accuracy of the symbol appearance probability may be inferior to an unlearned state. Not a few. However, according to the above configuration, when the partial area in the image corresponding to the encoding target symbol shifts to an area of a different image type in the image, the contents of the learning table are substantially initialized by the learning table changing unit. Therefore, the period required for optimizing the learning table for a new image type is shortened. As a result, it is possible to suppress a decrease in the compression rate in the boundary region between objects of different image types.

  Also, in this case, the initial table storage means stores a plurality of the initial tables for each image type, and the learning table change means determines that the image type detected by the image type detection means is the previous image type. When the image type is different from the image type detected by the type detection unit, the learning table is preferably rewritten based on the initial table corresponding to the image type of the partial area corresponding to the encoding target symbol.

  According to the above configuration, the learning table is rewritten by the learning table changing unit based on the initial table corresponding to the image type of the partial area corresponding to the encoding target data. Therefore, when the partial area in the image corresponding to the encoding target symbol shifts to an area of a different image type in the image, the learning table is rewritten with contents suitable for the image type of the transfer destination. The period required for optimization for the previous image type can be further shortened, and as a result, the compression rate can be further improved.

  Alternatively, the learning table storage unit stores the learning table for each image type, and the learning table change unit determines that the image type detected by the image type detection unit is previously detected by the image type detection unit. If the detected image type is different from the detected image type, the learning table referenced and updated by the encoding means may be switched to a learning table corresponding to the image type of the partial area corresponding to the encoding target symbol. Good.

  According to the above configuration, a learning table is prepared for each image type, and when the partial region in the image corresponding to the encoding target symbol is shifted to a region of a different image type in the image, the encoding means Since the learning table to be used is also switched to that for the destination image type, there is no need to re-optimize the learning table. Therefore, the compression rate can be further improved.

  In addition, the image data encoding device includes a drawing command receiving unit that receives a drawing command that instructs to place a desired type of object in a desired position in the image in order to create the image, and the drawing command receiving unit. Is further provided with image data creation means for creating the image data in accordance with the drawing command received by the image processing unit, and the image type detection means applies the symbol to be encoded based on the drawing command received by the drawing command reception means. It is preferable to detect the image type of the corresponding partial area.

  According to the above configuration, the determination of the image type by the image type detection unit is performed based on the drawing command. This drawing command is a command for creating image data encoded by the image data encoding device, and a desired object (for example, a text object, a graphics object, a bitmap object, etc.) is arranged at a desired position. This is an instruction to instruct. Since the image type detection means determines the image type of the partial area corresponding to the encoding target symbol based on the above drawing command, instead of using a statistical method from image data as in the past, the determination accuracy is high. improves. Furthermore, since no statistical method is used, processing necessary for the determination is reduced, and encoding can be performed efficiently. As described above, according to the above configuration, it is possible to easily and accurately detect which image type region in an image the data to be encoded is derived from.

  More specifically, the image data encoding device acquires the type of an object arranged in the image and a range occupied by the object in the image from the drawing command received by the drawing command receiving unit. And classifying the image into partial images for each image type based on the type and range of the acquired object, and region information indicating the image type of the divided partial image and the range occupied by the partial image in the image. A region information creating unit for creating the image information, and the image type detecting unit detects the image type of the partial region corresponding to the encoding target symbol based on the region information created by the region information creating unit. Can be mentioned.

  The image data encoding device adds region information to the encoded image data obtained by encoding all symbols constituting the image data. Preferably further means are provided.

  According to the above configuration, the area information necessary for decoding is added to the encoded image data, so it is not necessary for the user or the like to provide the area information, and the burden on the user is reduced. .

  The probability estimation table storage means stores a plurality of the probability estimation tables for each image type, and the image type detected by the image type detection means is detected by the image type detection means before. Table switching means for switching the probability estimation table referred to by the encoding means to a probability estimation table corresponding to the image type of the partial area corresponding to the encoding target symbol when the image type is different from the image type; It is preferable to provide.

  According to the above configuration, the probability estimation table indicating the update information of the learning table is prepared for each image type, and the partial region in the image corresponding to the encoding target symbol is shifted to a region of a different image type in the image. Accordingly, the probability estimation table used by the encoding means is also switched to that for the destination image type. As a result, when the partial region in the image corresponding to the encoding target symbol is shifted to a region of a different image type in the image, the learning table optimization method (learning method) is suitable for the destination image type. Therefore, the period required for the optimization of the learning table is shortened. Therefore, the compression rate can be further improved.

  The image data encoding device further includes data holding means for holding an arithmetic code obtained by sequentially encoding the encoding target symbols by the encoding means, and the encoded data holding means includes: When the image type detected by the image type detection unit is different from the image type previously detected by the image type detection unit, the arithmetic code is output and the arithmetic code held by itself is output. It is preferable to reset.

  According to the above configuration, when the learning table is changed, an arithmetic code obtained by encoding is output. Therefore, the decoding process at the changed part of the learning table is facilitated at the time of decoding.

  The encoding means may perform encoding according to the QM-Coder method. The image data encoding device further includes a data output means for outputting the encoded image data obtained by encoding the image data by the encoding means to the image forming apparatus. You may have. Each of the above means may be realized by a computer executing a program, and the drawing command receiving means may receive the drawing command from an application program via an operating system.

  By the way, the image data encoding apparatus may be realized by hardware or may be realized by causing a computer to execute a program. Specifically, the program according to the present invention is a program that causes a computer to operate as each unit of the image data encoding device, and the program is recorded on the recording medium according to the present invention.

  When these programs are executed by a computer, the computer operates as the image data encoding device. Therefore, even image data in which objects of a plurality of different image types are mixed can be encoded at a high compression rate.

  As described above, the image data encoding device according to the present invention is detected by the image type detection unit that detects the image type of the partial region in the image corresponding to the extracted encoding target symbol, and the image type detection unit. When the image type is different from the previously detected image type, there is provided a learning table changing means for changing the learning table stored in the learning table storage means.

  The image data encoding method according to the present invention includes an image type detection step in which the image type detection means detects an image type of a partial region in the image corresponding to the extracted encoding target symbol, and an image type detection step. A learning table changing step in which the learning table changing means changes the learning table stored in the storage means when the image type detected in the step is different from the image type previously detected in the image type detecting step. It is a configuration that includes.

  Therefore, as described above, even image data in which objects of a plurality of different image types are mixed can be encoded at a high compression rate.

  An image data encoding apparatus and a control method thereof according to an embodiment of the present invention will be described below with reference to FIGS. The image data encoding apparatus according to the present invention encodes image data by arithmetic encoding. In this embodiment, as an example, a case where the implementation form of arithmetic encoding is particularly QM-coder will be described. . In other words, the present embodiment is based on the ITU-T (International Telecommunication Union Telecommunication standardization sector) WHITE BOOK digital still image compression coding related recommendation (New Japan ITU Association) recommendation T.264. 82, an example applied to JBIG's QM-Coder.

  Regarding the image data encoding device of the present embodiment, in this specification, the configuration based on JBIG's QM-Coder and not related to the features of the present invention may be omitted as appropriate. For configurations not mentioned in this specification, ITU-T Recommendation T.30 is used unless a contradiction arises. 82. However, in the image data encoding apparatus according to the present invention, the implementation form of arithmetic encoding is not limited to JBIG's QM-Coder. For example, QM-Coder may be appropriately improved.

  JBIG's QM-Coder uses a method in which binary image data is modeled by a Markov information source, an encoding target symbol is predicted based on a pattern of surrounding symbol groups, and arithmetic coding is performed according to a prediction result. Yes. The JBIG compression uses a progressive method, but here, in order to simplify the explanation, the JBIG compression having only the lowest resolution layer is dealt with. For the sake of simplicity, the description of additional functions included in JBIG specifications such as AT (Adaptive Template) movement that changes the shape of the context will be omitted.

  First, an outline of arithmetic encoding performed by the image data encoding device 1 according to the present embodiment will be described. The image data encoding device 1 compresses binary image data by arithmetic encoding and outputs compressed image data (encoded image data) obtained thereby. The image data encoding device 1 functions as an encoding / compression unit in a printer driver of a personal computer (not shown), for example. This printer driver is realized by a CPU (Central Processing Unit) of a personal computer reading and executing a program stored in a storage unit (recording medium) (not shown) of the personal computer. The printer driver converts document data into image data by a drawing command in accordance with a user instruction to the personal computer, encodes the image data using the image data encoding device 1, and transmits the image data to the printer body. The printer main body that has received the encoded image data decodes the image data and performs printing.

  FIG. 2 is a diagram for explaining the concept of arithmetic compression. The image data encoding device 1 converts an input bit string (input symbol string), which is binary image data input from the outside, into a real number x between 0 and 1 in order to perform arithmetic encoding. More specifically, by dividing a half-open interval of 0 or more and less than 1 according to an input bit (input symbol), a half-open interval (interval of x or more and less than y) corresponding to the real number x is obtained. Let the lower limit value be the encoding result. As a result, the encoded data that is the encoding result is a single decimal with a large number of digits. Hereinafter, a half-open interval (a range in which the real number x can exist) corresponding to the real number x that is the encoding result is referred to as an interval.

  As shown in FIG. 2, the interval is clearly an area of 0 or more and less than 1 when 1 bit (1 symbol) has not yet been input. The image data encoding apparatus 1 reduces this interval according to a predetermined procedure according to the input bits. And the low number which is the lower limit (minimum value) of the interval after changing the interval according to all the input bits is the encoding result.

  In the image data encoding device 1, the interval is stored as a combination of a low number that is the lower limit value of the interval and the size of the interval (difference between the lower limit value and the upper limit value). More specifically, the image data encoding apparatus 1 converts the current input bit into an input bit string up to the previous input bit, more specifically a pattern of peripheral bit groups (a plurality of inputs corresponding to peripheral pixels of the input pixel). Prediction from bit value combinations). Hereinafter, this peripheral bit pattern is referred to as a context.

  The image data encoding device 1 has a function of estimating a value that is highly likely to appear in the encoding target bit and the appearance probability of the value from the above-described context. As a result, it is determined from the state of surrounding pixels which value (symbol) of 0 or 1 is likely to appear as an input bit.

  Here, a value (symbol) having a higher probability of appearing in the input bit among 0 and 1 values that can be taken by the input bit is referred to as a dominant probability symbol (hereinafter also referred to as “MPS”). The remaining one (the value with the lower probability of appearing in the input bit) is referred to as an inferior probability symbol (hereinafter also referred to as “LPS”). That is, the image data encoding device 1 has a function of estimating the appearance probability of the dominant probability symbol and the inferior probability symbol.

  Then, the image data encoding device 1 divides the interval into two regions according to the ratio between the appearance probability of the dominant probability symbol and the appearance probability of the inferior probability symbol, and one region is an MPS interval corresponding to the MPS appearance probability, The other area is set as an LPS interval corresponding to the appearance probability of LPS. Here, of the two divided areas, an area closer to 0 is an MPS interval, and an area far from 0 is an LPS interval.

  When the input bit string (input symbol string) is encoded, the input bit (input symbol) and the dominant probability symbol (MPS) are compared according to the input of each bit (symbol). If the input bit matches the dominant probability symbol (MPS), the interval is changed to remove the LPS interval from the current interval. On the other hand, if the input bit and the MPS do not match, the interval is changed so as to remove the MPS interval from the current interval.

  As an example, an encoding process when the input bit string (input symbol string) is “0001” and the dominant probability symbol (MPS) is “0101” will be described with reference to FIG. In this example, since the input value of the second bit (second input symbol from the front) is an inferior probability symbol, when the input is encoded, the LPS interval is selected as a new interval after encoding. That is, encoding is performed by changing the interval so as to remove the MPS interval from the current interval. On the other hand, since the input values other than the second bit (the first, third, and fourth symbols from the front) are dominant probability symbols, when the input is encoded, the MPS interval is set as the encoded interval. Selected. That is, encoding is performed by changing the interval so as to remove the LPS interval from the current interval.

  Here, a low number of intervals when processing for all input bit strings is completed is a real number x, and this real number x is an encoding result (code data). If the input bit matches the expected input bit (dominance probability symbol) by the above operation, the low number of intervals does not change. For this reason, there is no change in the output value (encoding result), and there is no need to output a new bit, so that encoding can be performed using this. If all input bits are dominant probability symbols, the encoding result x is 0.

  In performing the operation as described above, the image data encoding apparatus 1 that imitates the operation of the QM-Coder further performs the following procedure. That is, as the above arithmetic coding process is performed, the interval size gradually decreases. For this reason, it will eventually become impossible to express with the fixed precision of the computer. Therefore, a process called normalization processing is performed in which the size of the interval is doubled until the size of the interval becomes 0.5 or more.

In the image data encoding device 1, the size of the interval is held (stored) in a register A (hereinafter referred to as “Areg”) 27a in a register 27 described later, and the low number of intervals is stored in a register 27 described later. Is stored (stored) in a register C (hereinafter referred to as “Creg”) 27b. In the present embodiment, the interval, the actual value of 2 ¹⁶ times the value, i.e., expressed by a value in the range of 0X0～0x10000. In the following, the hexadecimal representation of the value (value preceded by a "0x") is a 2 ¹⁶ times the value of the interval. Accordingly, the actual interval value is obtained by dividing the hexadecimal display value by 2 ¹⁶ . In this case, the size of the interval normalized so as to be 0.5 (“0x8000”) or more and 1 (“0x10000”) or less is stored in the Areg 26a.

  Further, every time the value (interval size) stored in the Areg 26a is doubled by the normalization process, the value (low number of intervals) stored in the Creg 26b is also doubled. For this reason, the influence on the output caused by doubling the value stored in the Areg 26a is canceled (cancelled). In order to prevent overflow of the value of Creg 26b, upper bytes are taken out in a timely manner and output as an encoding result. It should be noted that a method for solving the carryover that occurs at this time is described in ITU-T TT. However, since it is not essential for understanding the contents of the present invention, the description thereof is omitted here.

Further, the image data encoding device 1 performs approximation as follows. To calculate the LPS interval,
(Interval size) x (LPS appearance probability)
This operation is necessary. In the image data encoding device 1, the value of this LPS interval is approximated by a statistically obtained estimated value (estimated LPS interval; hereinafter, abbreviated as “LSZ” as appropriate). In this approximation, it is assumed that the interval size is 1 without considering that the interval size at that time is not always 1 ("0x10000"). Is approximated by the estimated value (LSZ) of the appearance probability of LPS obtained in step (1). Therefore, the proportion of the estimated LPS interval in the whole interval changes according to the size of the interval. However, since the interval is normalized so that the size is 0.5 or more and 1 or less, the error is not so large.

  That is, the above approximation is performed assuming that the size of the entire interval is the sum of the appearance probability of 0 and the appearance probability of 1, so that the size of the entire interval should be 1 originally. is there. However, for convenience of calculation, the process of setting the entire interval to 1 is not performed, and the entire interval needs to be within a range from 0.5 to 1.0. For this reason, if the size of the interval is small, when the MPS interval is obtained by subtracting the LPS interval from the interval, the size of the LPS interval may be larger than the size of the MPS interval.

  In summary, in the image data encoding device 1, when MPS is given as an input bit (input symbol), LSZ is subtracted from the interval, and the resulting interval is set as a new interval. Also, the image data encoding device 1 changes the interval to LSZ when LPS is given as an input bit. By this operation, the input bits are encoded. The image data encoding apparatus 1 performs normalization when the interval becomes 0.5 or less as a result of this operation. When normalization is performed, the value of LSZ is updated. In addition, the LPS interval may become larger than the MPS interval due to the above encoding. At this time, the MPS interval and the LPS interval are exchanged.

  FIG. 1 is a functional block diagram showing the configuration of the image data encoding device 1 of the present embodiment. As shown in FIG. 1, the image data encoding device 1 includes a drawing command receiving unit (drawing command receiving unit) 11, an area information generating unit (region information generating unit) 12, and an image data generating unit (image data generating unit) 13. , Context generation unit (extraction unit) 21, region determination unit (image type detection unit) 22, table change unit (learning table change unit, table switching unit) 23, acquisition unit 24 (encoding unit), arithmetic encoding unit ( Encoding means, loop means 26, register (data holding means) 27, area information adding section (area information adding means, data output means) 28 and recording section (learning table storage means, probability estimation table storage means, initial table storage) Means) 31. In the present embodiment, each of these units is realized by execution of a program code stored in a recording medium such as a ROM or a RAM by an arithmetic unit such as a CPU (not shown) provided in the image data encoding device 1. That is, the image data encoding device 1 is a computer that includes an arithmetic unit such as a CPU and a storage device such as a RAM as hardware resources.

  The drawing command receiving unit 11 receives a drawing command for creating an image. For example, (1) an instruction to draw a text object on the canvas, (2) an instruction to draw a graphics object on the canvas, and (3) a designation on the canvas A command for instructing to draw the bitmap object is included.

  The instruction (1) is an instruction for arranging desired text in a desired font stored in a font storage unit (not shown) on the canvas. The above instruction (2) is an instruction for arranging, for example, a straight line, a curve, a circle, a polygon, or a combination thereof, defined by a function on the canvas. The command (3) is a command for placing a bitmap image such as a photograph expressed by halftone dots or error diffusion on the canvas.

  When the encoding engine of the image data encoding device 1 is implemented as a printer driver of a computer, for example, the drawing command is input from the application program to the image data encoding device 1 via the operating system. For example, when the operating system is Windows (registered trademark), the application program calls a GDI (Graphic Device Interface) function of the operating system when performing print output. Then, the called GDI function further calls a DDI (Device Driver Interface) function of the printer driver. As a result, the called DDI function creates a bitmap image in which the designated object is drawn and outputs it to the printer. Here, the above-mentioned DDI function call instruction is an example of the above-described drawing instruction.

  The name of the DDI function varies depending on the operating system, but in the case of Windows (registered trademark), any name other than “DrvEnableDriver ()” can be freely assigned. For example, “DrvBitBlt ()” or “DrvTextOut ()”. That is, when Windows (registered trademark) is adopted as the operating system, a drawing function that can be received by the drawing command receiving unit 11 and a call command thereof can be freely created. In addition to the above function name indicating the type of drawing object, the DDI function call instruction includes information specifying the size of the drawing object and the arrangement on the image as arguments.

  The drawing command receiving unit 11 can receive a plurality of the drawing commands described above. Accordingly, the image data creation unit 13 described later can create image data of an image in which a plurality of different types of objects are arranged on the canvas in accordance with a plurality of drawing commands. The drawing command received by the drawing command receiving unit 11 is sent to the area information creating unit 12 and the image data creating unit 13.

  The area information creating unit 12 creates area information from the drawing command received from the drawing command receiving unit 11. This area information is information indicating where and what type (image type) of image (partial image) is included in the image created by the image data creation unit 13. The area information created by the area information creating unit 12 is stored in the recording unit 31.

  In order to create this area information, the area information creating unit 12 first detects the type of object arranged on the image and the range occupied by the object on the image from the drawing command. Here, the type of the object included in the image is determined based on the function name included in the drawing command. The range that the object occupies on the image is recognized based on the function argument included in the rendering command.

  In the present embodiment, the area information creation unit 12 classifies each object arranged on the image into any one of the above-described text object, graphics object, and bitmap object. Then, the created image is classified based on the type of object.

  FIG. 3 is a diagram illustrating an example of classification by the region information creation unit 12. As illustrated in FIG. 3, the region information creation unit 12 divides an image into a text region including a text object, a graphics region including a graphics object, and a bitmap region including a bitmap object. . In the present embodiment, as shown in FIG. 3, image segmentation by the region information creation unit 12 is performed only in the vertical direction and not in the horizontal direction. That is, the segmentation of the image by the region information creation unit 12 is performed by the (horizontal) line of the image.

  Then, the region information creation unit 12 creates region information indicating the positions of the upper and lower ends of each region (the range occupied by the partial image of each image type) and the type of the region. With this area information, it is possible to recognize which area of the text area, the graphics area, and the bitmap area an arbitrary line (and pixels included therein) in the image is included.

  If objects of different types overlap, priority is given to bitmap, graphics, and text. For example, a line including both a bitmap object and a graphic object is recognized as a bitmap area. A line where no object is arranged may be recognized as any region.

  The image data creation unit 13 draws an image in accordance with the drawing command received from the drawing command reception unit 11 and creates bitmap image data indicating the image. This bitmap image data is data to be subjected to arithmetic coding. In the present embodiment, the image drawn by the image data creation unit 13 is a binary (monochrome two gradation) bitmap image. However, the present invention is not limited to monochrome two gradations. The bitmap image data created by the image data creation unit 13 is sent to the context generation unit 21.

  From the bitmap image data received from the image data creation unit 13, the context generation unit 21 includes a pixel (encoding target symbol) to be encoded and a context (peripheral symbol) indicating a pattern of a plurality of surrounding pixels. Is extracted. FIG. 4 is a diagram illustrating a relationship between a pixel to be encoded and a context. As described above, in the image data encoding apparatus 1 that imitates the QM-coder, not only the input bit string up to the previous input signal but also the patterns (contexts) of a plurality of surrounding pixels are used for prediction. It has become. In other words, the image data encoding apparatus 1 estimates the appearance probability of the dominant probability symbol and the inferior symbol based on the combination (context) of states that can be taken by a plurality of neighboring pixels at the time of prediction.

  Here, when the target pixel having the current input bit is selected, as the peripheral pixels of this encoding target pixel, the same as the target pixel among the pixels processed before the target pixel, as shown in FIG. A plurality of pixels existing in a line and a line adjacent to the line (upper line in FIG. 3) are selected as peripheral pixels. In the example of FIG. 3, the hatched portion is a pixel to be encoded, and a pattern indicated by 10 pixels around it is a context.

  In the present embodiment, the number of pixels in the context is 10, and the value (symbol) of each pixel has a value of 0 or 1, so the context can take 1024 patterns. In the JBIG compression, as the context of the lowest resolution layer, a 3-line template is also prepared in addition to the 2-line template as shown in FIG. 3, and this can also be used. Here, a two-line template is used.

  The context generation unit 21 sends the extracted pixel value (encoding target symbol) and context (neighboring symbols) to the acquisition unit 24. Further, when receiving the pixel increment instruction from the arithmetic encoding unit 26, the context generation unit 21 extracts the next target pixel and context from the input bitmap image data. Further, in the present embodiment, when the pixel to be encoded moves to the next line according to the pixel increment instruction, the context generation unit 21 obtains position information indicating the position of the new line to be transferred in the image. This is sent to the area determination unit 22.

  The acquisition unit 24 acquires a dominant probability symbol (MPS) corresponding to the context received from the context generation unit 21 from the learning table 32 stored in the recording unit 31. That is, the acquisition unit 24 has a function of estimating the value of the encoding target pixel from the context. Furthermore, the acquisition unit 24 acquires the index ST of the probability estimation value of the inferior probability symbol corresponding to the context received from the context generation unit 21 from the learning table 32. The acquired MPS and ST are sent to the arithmetic encoding unit 26.

  Here, the learning table 32 stored in the recording unit 31 will be described. FIG. 5 is a diagram showing the contents of the learning table 32. As shown in FIG. 5, in the learning table 32, a context, an index ST for specifying the appearance probability (LSZ) of an inferior probability symbol estimated from the context, and a dominant probability symbol MPS estimated from the context, The correspondence relationship is shown. That is, the recording unit 31 that stores the learning table 32 stores a context, ST, and MPS in association with each other.

  The learning table 32 is optimized as the pixels included in the image are sequentially encoded. Specifically, the index ST and MPS are updated by the arithmetic encoding unit 26 in the data included in the learning table 32 as the pixels are encoded. In the present embodiment, the learning table 32 is created using the initial learning table 33 stored in the recording unit 31 as a template. In other words, in the initial state, the learning table 32 holds the same contents as the initial learning table 33, but is subsequently optimized as the pixels are encoded.

  The dominant probability symbol shown in the learning table 32 takes a value of 0 or 1. Further, the index ST of LSZ takes any value from 0 to 112 in JBIG. The LSZ index ST is a value that has appeared at the position of the pixel of interest so far for each pattern (context) of a value (symbol) of 0 or 1 that appears in the neighboring 10 pixels in the JBIG probability estimation table 34 described later ( This is used to specify the appearance probability of a value (symbol) that will appear at the target pixel from the history of the symbol.

  The arithmetic encoding unit 26 generates a real number x, which is an arithmetic code, as encoded data from the encoding target pixel value, MPS, and index ST received from the acquisition unit 24. In the present embodiment, the operation of arithmetic coding by the arithmetic coding unit 26 is the same as that of the conventional QM-Coder. The arithmetic encoding unit 26 includes a preprocessing unit 26a, an MPS processing unit 26b, an LPS processing unit 26c, and a RENORM processing unit 26d.

  The preprocessing unit 26a compares the value of the encoding target pixel (coding target symbol) received as the encoding target symbol from the acquisition unit 24 with the MPS, and determines whether or not the encoding target symbol matches the MPS. Determine. Here, the MPS processing unit 26b functions when the encoding target symbol matches MPS, and the LPS processing unit 26c functions when they do not match. That is, the MPS processing unit 26b is for encoding the dominant probability symbol, and the LPS processing unit 26c is for encoding the inferior probability symbol. The RENORM processing unit 26d is for performing the normalization process described above.

  The MPS processing unit 26b and the LPS processing unit 26c encode the symbol based on the symbol received from the acquisition unit 24 and the index ST, and obtain a real number x. When performing the arithmetic coding, the MPS processing unit 26b or the LPS processing unit 26c first refers to the probability estimation table 34 stored in the recording unit 31 to estimate the appearance probability of the inferior probability symbol corresponding to the index ST. Get LSZ.

  Here, the probability estimation table 34 will be described. FIG. 6 is a diagram illustrating the contents of one of the plurality of probability estimation tables 34... Stored in the recording unit 31. As shown in FIG. 6, the probability estimation table 34 shows the correspondence between the index ST and the appearance probability estimated value LSZ of the inferior probability symbol. In other words, ST and LSZ are stored in association with each other in the recording unit 31 that stores the probability estimation table 34. As described above, the LSZ is used to change the interval.

  In addition, the probability estimation table 34 shows the correspondence between the index ST and the updated contents of the learning table 32. Specifically, the index ST is associated with NLPS, NMPS, and SWITCH.

  Here, NLPS means the value of the next index ST associated with the context when the LPS processing unit 26c encodes the inferior probability symbol. That is, when the encoding target symbol is an inferior probability symbol, among the STs included in the learning table 32, the ST corresponding to the context extracted by the context generation unit 21 is overwritten by the NLPS.

  On the other hand, NMPS means the value of the next index ST associated with the context when the MPS processing unit 26b encodes the dominant probability symbol. That is, when the encoding target symbol is a dominant probability symbol, among the STs included in the learning table 32, the ST corresponding to the context extracted by the context generation unit 21 is overwritten by the NMPS.

  SWITCH is a flag indicating whether or not to replace the dominant probability symbol and the inferior probability symbol. In the case of 1, SWITCH indicates that the symbol is replaced. That is, when the SWITCH associated with the index ST received together with the encoding target symbol is 1, the MPS processing unit 26b and the LPS processing unit 26c invert the MPS value in the learning table 32 (from 0 to 1 or from 1) Change to 0).

  As described above, the recording unit 31 that stores the probability estimation table 34 stores ST, LSZ, NLPS, MPS, and SWITCH in association with each other. By updating the contents of the learning table 32 based on the update information, the appearance probability value of the inferior probability symbol estimated from the context is updated to take into account the encoded symbol.

  In the present embodiment, the probability estimation table 34 is prepared for each of the three image types. That is, a probability estimation table 34 for three regions for a text region, a graphics region, and a bitmap region is stored in the recording unit 31, and the arithmetic encoding unit 26 includes a line including an encoding target pixel. The probability estimation table 34 corresponding to the region to which the

  Note that the probability estimation table 34 takes into account a dither pattern if it is for a bitmap area, for example, while it is conscious of a line drawing if it is for a graphics area. The probability estimation table 34 suitable for each region may be created by experimentally encoding a number of images of each image type and obtaining a table that statistically improves the compression rate.

  The region determination unit 22 receives the pixel received from the context generation unit 21 as to which of the text region, the graphics region, and the bitmap region the pixel to be encoded extracted by the context generation unit 21 belongs. And the area information 35 stored in the recording unit 31. This determination result is held in the area determination unit 22. Then, the region determination unit 22 compares the determination result for the current encoding target pixel with the determination result for the previous encoding target pixel that is held, and if the two are different, the table change unit 23 To change the table.

  The table changing unit 23 resets the learning table 32 stored in the recording unit 31 and switches the probability estimation table 34 referred to by the arithmetic coding unit 26 in accordance with an instruction from the region determination unit 22. .

  The learning table 32 is reset by overwriting the data in the learning table 32 based on the contents of the initial learning table 33 in an unoptimized state. Thereby, in this embodiment, the learning table 32 is reset to the unlearned initial state. On the other hand, switching of the probability estimation table 34 referred to by the arithmetic encoding unit 26 is performed by causing the arithmetic encoding unit 26 to refer to the probability estimation table 34 corresponding to the region to which the pixel to be encoded belongs.

  When the region determination unit 22 and the table change unit 23 shift the pixels to be encoded to regions of different image types, the learning table is reset and the probability estimation referred to by the arithmetic encoding unit 26 The table 34 is switched to the one suitable for the new area.

  The register 27 holds the arithmetic code (real number x) generated by the arithmetic encoding unit 26. As described above, the register 27 has the Areg 27a and the Creg 27b, and the size of the interval is recorded in the Areg 27a, and the low number of intervals that are the real number x is recorded in the Creg 27b.

  The region information adding unit 28 adds the region information created by the region information creating unit 12 to the encoded image data when all the image data is encoded, and can be used for decoding. It is what you want to do.

  Each of the blocks included in the image data encoding device 1 can be configured by hardware logic, but in the present embodiment, it is realized by software using a CPU as follows.

  That is, the image data encoding apparatus 1 includes a CPU (central processing unit) that executes instructions of a control program that realizes each function, a ROM (read only memory) that stores the program, and a RAM (random access) that expands the program. memory), a storage device (recording medium) such as a memory for storing the program and various data. An object of the present invention is a recording in which a program code (execution format program, intermediate code program, source program) of a control program of the image data encoding device 1 which is software for realizing the above-described functions is recorded so as to be readable by a computer. This can also be achieved by supplying a medium to the image data encoding apparatus 1 and reading and executing the program code recorded on the recording medium by the computer (or CPU or MPU).

  Examples of the recording medium include a tape system such as a magnetic tape and a cassette tape, a magnetic disk such as a floppy (registered trademark) disk / hard disk, and an optical disk such as a CD-ROM / MO / MD / DVD / CD-R. Card system such as IC card, IC card (including memory card) / optical card, or semiconductor memory system such as mask ROM / EPROM / EEPROM / flash ROM.

  Further, the image data encoding device 1 may be configured to be connectable to a communication network, and the program code may be supplied via the communication network. The communication network is not particularly limited. For example, the Internet, intranet, extranet, LAN, ISDN, VAN, CATV communication network, virtual private network, telephone line network, mobile communication network, satellite communication. A net or the like is available. Also, the transmission medium constituting the communication network is not particularly limited. For example, even in the case of wired such as IEEE 1394, USB, power line carrier, cable TV line, telephone line, ADSL line, etc., infrared rays such as IrDA and remote control, Bluetooth ( (Registered trademark), 802.11 wireless, HDR, mobile phone network, satellite line, terrestrial digital network, and the like can also be used. The present invention can also be realized in the form of a computer data signal embedded in a carrier wave in which the program code is embodied by electronic transmission.

  Next, the operation of the image data encoding device 1 of this embodiment will be described. 7 to 9 are flowcharts showing processing by the image data encoding device 1 of the present embodiment.

  First, the drawing command reception unit 11 of the image data encoding device 1 receives the above-described drawing command from an application program or the like (S11). Then, the drawing command reception unit 11 sends the received drawing command to the area information creation unit 12 and the image data creation unit 13.

  Next, the region information creation unit 12 classifies the image created by the image data creation unit 13 into partial images for each image type as shown in FIG. Area information indicating the range of the partial image in the image is created (S12). Then, the area information creation unit 12 stores the created area information in the recording unit 31.

  Next, the image data creating unit 13 creates bitmap image data of a binary image according to the drawing command received from the drawing command receiving unit 11 (S13). The image corresponding to this image data is, for example, an image shown in a portion surrounded by a frame in FIG. Then, the image data creation unit 13 sends the created image data to the context generation unit 21.

  When receiving the image data, the context generation unit 21 divides the image (bitmap image) corresponding to the received image data into at least one stripe (S14). This stripe is a unit for performing compression processing, and is composed of one or more horizontal lines in the image. Then, the context generation unit 21 selects a stripe included in the pixel to be compressed. The selection order of the stripes is not particularly limited. For example, the stripes are sequentially selected from the top to the bottom of the image. When the processing for all stripes is completed by the loop (LOOP_A) for the stripes in S14 to S34, the compression of the entire image is completed.

  Subsequently, the context generation unit 21 selects a pixel (target pixel) to be compressed (S15). The selection of the target pixel is not particularly limited, but is sequentially performed from the left end to the right end of the stripe, for example. When the target pixel reaches the end of the line, the process proceeds to the next line. In addition, when the compression for the target pixel is performed in the loop (LOOP_B) of S15 to S32 and the processing of all the pixels included in the stripe is completed, the processing of one stripe is completed. Then, the context generation unit 21 sends the position information of the target pixel to the region determination unit 22.

  Next, the region determination unit 22 determines whether the target pixel is at the head of the line based on the position information of the target pixel received from the context generation unit 21 (S16). If the target pixel is not the head of the line, the process proceeds to S22. On the other hand, if the target pixel is the head of the line, the process proceeds to S17.

  In S <b> 17, the region determination unit 22 determines whether the target pixel is included in the above-described text region, graphics region, or bitmap region, the position information of the target image received from the context generation unit 21, and the recording unit. This is determined based on the area information stored in the area 31. Subsequently, the region determination unit 22 compares the current region determination result with the previous region determination result held and determines whether the two region determination results are different (S18).

  Here, if the two region determination results are different, it is determined that the learning table 32 needs to be reset and the probability estimation table 34 needs to be switched, the current region determination result is retained, and the process proceeds to S19. On the other hand, if the two area determination results match, the current area determination result is held and the process proceeds to S22.

  In step S <b> 19, the table changing unit 23 switches the probability estimation table 34 referred to by the arithmetic encoding unit 26 described later in accordance with the instruction from the region determination unit 22. In the present embodiment, as described above, the probability estimation tables suitable for each of the three areas are stored in the recording unit 31, and from these tables, the table changing unit 23 is operated by the area determining unit 22. Based on the region determination result, the probability estimation table 34 suitable for the region including the target pixel is selected. Then, the table changing unit 23 sets the selected table 34 as the probability estimation table 34 referred to by the arithmetic encoding unit 26. By this processing, the probability estimation table referred to by the arithmetic coding unit in S25, S28, S29, and S31 described later is changed to the one selected by the table changing unit 23.

  Subsequently, the table changing unit 23 resets the learning table 32 of the recording unit 31 (S20). In the present embodiment, the learning table 32 is reset by overwriting the contents of the learning table 32 with the contents of the initial learning table 33 stored in the recording unit 31 as described above. Thereby, the content of the learning table 32 returns to the unlearned initial state.

  Next, the register 27 performs a flash process for outputting the encoded data (arithmetic code) held in the Creg 27b to the area information adding unit (S21). Thereby, the contents of Areg 27a and Creg 27b are reset.

  In S <b> 22, the context generation unit 21 reads the value of the target pixel as an encoding target symbol and reads the values of the peripheral pixel group to generate a context. Then, the encoding target symbol and the context are sent to the acquisition unit 24.

  Next, the acquisition unit 24 acquires, from the learning table 32 of the recording unit 31, based on the context received from the context generation unit 21, the index ST of the probability estimation values of the dominant probability symbol (MPS) and the inferior probability symbol ( S23). Specifically, the acquisition unit 24 acquires the MPS and ST corresponding to the received context in the learning table 32. Then, the acquisition unit 24 sends the encoding target symbol and the acquired MPS and ST to the arithmetic encoding unit 26.

  Next, the preprocessing unit 26a of the arithmetic encoding unit 26 compares the received encoding target symbol with the MPS, and determines whether or not the encoding target symbol is MPS (S24). If the encoding target symbol is MPS, the process proceeds to S25, and the MPS processing unit 26b performs arithmetic encoding (CODE_MPS: S25 to S28). On the other hand, when the value of the target pixel is not MPS but LPS, the process proceeds to S29, and the MPS processing unit 26b performs arithmetic coding (CODE_LPS: S29 to S31).

  In S25, the MPS processing unit 26b refers to the probability estimation table 34 of the recording unit 31, and acquires the probability estimation value (LSZ) of the inferior probability symbol corresponding to the index ST received from the acquisition unit 24. Further, the MPS processing unit 26b acquires information on the current interval from the Areg 27a and the Creg 27b of the register 27. Then, the MPS processing unit 26b divides the current interval based on the LSZ, and stores information on the new interval obtained by the division in the Areg 27a and the Creg 27b of the register 27 (S25). As a result, the register 27 stores arithmetic code data obtained by encoding pixel data including the value (dominance probability symbol) of the pixel selected in S15.

  Next, the MPS processing unit 26b determines whether or not the value of the Areg 27a is less than 0.5 (0x8000) (S26). If the value of Areg 27a is less than 0.5, the process proceeds to S27 and normalization processing is performed. On the other hand, when the value of Areg 27a is 0.5 or more, the process proceeds to S32.

  In S27, the RENORM processing unit 26d performs normalization processing on the Areg 27a and Creg 27b of the register 27 in accordance with the instruction of the MPS processing unit 26b (S27), and the encoded data is transferred to the register 27 as necessary. The region information adding unit 28 outputs the information.

  Next, the MPS processing unit 26 b refers to the probability estimation table 34 of the recording unit 31 and acquires the NMPS and SWITCH corresponding to the index ST received from the acquisition unit 24. Then, the MPS processing unit 26b updates the index ST corresponding to the context generated in S22 in the learning table 32 based on the acquired NMPS (S28). At this time, the MPS of the learning table 32 is also inverted according to SWITCH. As a result, the value of LSZ derived from the context by the learning table 32 is updated to take into account the encoding target symbol. Then, the process proceeds to S32.

  On the other hand, in S29, the LPS processing unit 26c refers to the probability estimation table 34 of the recording unit 31, and acquires the probability estimation value (LSZ) of the inferior probability symbol corresponding to the index ST received from the acquisition unit 24. Further, the LPS processing unit 26c acquires information on the current interval from the Areg 27a and the Creg 27b of the register 27. Then, the LPS processing unit 26c divides the current interval based on the LSZ, and stores information on the new interval obtained by the division in the Areg 27a and the Creg 27b of the register 27. As a result, the register 27 stores arithmetic code data obtained by encoding pixel data including the value of the pixel selected in S15 (inferior probability symbol).

  Next, the LPS processing unit 26 c refers to the probability estimation table 34 of the recording unit 31 and acquires the NLPS and SWITCH corresponding to the index ST received from the acquisition unit 24. Then, the LPS processing unit 26c inverts the MPS corresponding to the context generated in S22 in the learning table 32 according to the acquired SWITCH (S30).

  Subsequently, the LPS processing unit 26c updates the index ST corresponding to the context generated in S22 in the learning table 32 based on the acquired NLPS (S31). As a result, the value of LSZ derived from the context by the learning table 32 is updated to take into account the encoding target symbol.

  In S32, the MPS processing unit 26b or the LPS processing unit 26c instructs the context generation unit 21 to increment the pixel, and the process returns to S151. This completes the pixel-by-pixel encoding process. When this encoding process is performed for all the pixels included in the stripe selected in S14, the process proceeds to S33, and the register 27 performs the flash process of the encoded data. Then, the process returns to S14. This completes the stripe unit encoding process (S34).

  When this encoding process is performed for all the stripes included in the image, the region information adding unit 28 stores all of the encoded image data expressed by the arithmetic code. Finally, the area information adding unit 28 reads the area information 35 from the recording unit 31 and adds it to the encoded image data (S35), and sends the obtained data to an image forming apparatus (not shown) via the interface. Output and complete the process.

  As described above, the image data encoding device 1 according to the present embodiment includes the learning table 32 indicating the correspondence between the context that is the pattern of the peripheral symbols and the index ST for specifying the LSZ, and the index ST. A recording unit 31 that stores a probability estimation table 34 showing the correspondence between LSZ and NLPS, NMPS, and SWICTH, a context generation unit 21 that extracts the value and context of a pixel to be encoded from image data, and context generation The index ST corresponding to the context extracted by the unit 21 is acquired from the learning table 32, the LSZ and NLPS or NMPS corresponding to the acquired index ST are acquired from the probability estimation table 34, and the encoding target is based on the acquired LSZ. When the pixel value is encoded by arithmetic encoding Moni, and an arithmetic coding unit 26 to update the index ST included in the learning table 32 based on the NLPS or NMPS according to a result of comparison between the value of the coded pixel and MPS. In addition, the arithmetic encoding unit 26 outputs a pixel increment instruction in order to repeatedly execute the processing by each unit including itself.

  As a result, the image data encoding apparatus 1 uses the values of the pixels constituting the image as encoding target symbols, and sequentially encodes the encoding target symbols by arithmetic encoding based on the LSZ estimated from the pattern of the surrounding symbols. It becomes the composition which becomes.

  Here, the image data encoding device 1 of the present embodiment further includes an area determination unit 22 that detects an image type of an area including the encoding target pixel extracted by the context generation unit 21, and an area determination unit 22. A table changing unit 23 is provided for changing the learning table 32 when the detected image type is different from the previously detected image type.

  As a result, when the encoding target pixel shifts to a region of a different image type, the contents of the learning table 32 are changed by the table changing unit 23. Therefore, the learning table 32 optimized for other image types is used. Thus, the value of the encoding target pixel is not encoded. Therefore, it can suppress that a compression rate falls.

  Note that the image data encoding device 1 of the present embodiment is configured to reset (initialize) the learning table 32 as a method of changing the learning table 32 described above. That is, the recording unit 31 of the image data encoding device 1 of the present embodiment stores an initial learning table 33 that serves as a template for the learning table 32, and the table changing unit 23 is an area of image types with different encoding target pixels. In this case, the learning table 32 is reset based on the initial learning table 33.

  As a result, when the encoding target pixel shifts to a region of a different image type, the contents of the learning table 32 optimized for other image types are reset, so that the learning table 32 is used for a certain image type. Even after the optimization, even if the encoding target pixel shifts to a region having a different image type, it is possible to suppress a reduction in the compression rate.

  In the present embodiment, as described above, a single initial learning table 33 is prepared in the recording unit 31 and the learning table 32 is reset by the initial learning table 33. The initial learning table 33... Is prepared for each image type in the recording unit 31, and the table changing unit 23 calculates the arithmetic code when the encoding target pixel is shifted to a different image type region. The learning table 32 used by the conversion unit 26 may be reset based on the initial learning table 33 corresponding to the image type of the region including the encoding target pixel. That is, in this case, as shown in FIG. 10, the recording unit 31 stores three initial learning tables 33a to 33c for the text area, the graphics area, and the bitmap area, and the encoding target pixel belongs to the recording area 31. The learning table 32 is reset based on the area initial learning tables 33a to 33c.

  As a result, when the encoding target pixel shifts to a region having a different image type, the learning table 32 is rewritten with contents suitable for the image type of the transfer destination region, so that the period until the learning table 32 is optimized Is shortened. As a result, the compression rate can be improved.

  Note that the initial learning table 33 takes into account a dither pattern if it is for, for example, a bitmap area, while being aware of a line drawing if it is for a graphics area. The initial learning table 33 suitable for each region may be created by experimentally coding a number of images of each image type and obtaining a table that has a short period until optimization statistically.

  Moreover, the following is mentioned as a modification of the change method of the learning table 32. FIG. That is, in the modification, the learning table 32 itself is prepared for each image type in the recording unit 31 as shown in FIG. 11 instead of the initial learning table 33 that is a template of the learning table 32. That is, three learning tables 32a to 32c for the text area, the graphics area, and the bitmap area are stored in the recording unit 31. Then, the table changing unit 23 changes the learning table 32 used by the arithmetic coding unit 26 according to the image type of the region including the encoding target pixel when the encoding target pixel shifts to a region of a different image type. The learning table 32 is switched.

  As a result, when the encoding target pixel shifts to a region having a different image type, the learning table 32 is switched to one corresponding to the image type of the transfer destination region, and the text region is encoded when the text region pixel is encoded. The learning table 32a for the graphics area is used. When the pixels in the graphics area are encoded, the learning table 32b for the graphics area is used. When the pixels in the bitmap area are encoded, the learning table 32c for the bitmap area is used. Will be used. Therefore, it is not necessary to optimize the learning table 32 again when the encoding target pixel shifts to a region having a different image type. As a result, the compression rate can be further improved.

  In this embodiment, the table changing unit 23 always changes (resets) the learning table 32 when the encoding target pixel shifts to a region having a different image type. However, the present invention is not limited to this. Instead, the learning table may be changed (reset) when the encoding target pixel shifts to a region having a different image type and a predetermined condition is satisfied. In this case, for example, when the encoding target pixel shifts from the graphics area to the text area, and when the encoding target pixel shifts from the text area to the graphics area, the learning table is not changed (reset), and the encoding target pixel does not change. The learning table may be changed (reset) only when shifting from the bitmap area to another area and when shifting from another area to the bitmap area.

  In the present embodiment, the image type of the area including the encoding target pixel is detected as follows. That is, the image data encoding device 1 includes a drawing command receiving unit 11 that receives a drawing command for instructing to place an object of a desired type in a desired position in the image, and a drawing command receiving unit 11. Is further provided with an image data creation unit 13 for creating image data in accordance with the drawing command received by the user. Then, the region determination unit 22 detects the image type of the region including the encoding target pixel based on the drawing command received by the drawing command receiving unit 11.

  More specifically, the image data encoding device 1 acquires information on the type of object arranged in the image and the range occupied by the object in the image from the drawing command received by the drawing command receiving unit 11. The image is divided into a text area, a graphics area, and a bitmap area based on the acquired object type and range information, and the image type of each divided area and the range of the area in the image are indicated. An area information creation unit 12 for creating area information is further provided. Then, the region determination unit 22 detects the image type of the region including the encoding target pixel based on the region information created by the region information creation unit 12.

  As described above, in the image data encoding device 1 according to the present embodiment, each of the images in the image is based on the type of the object instructed by the drawing command for creating the image data and the range occupied by the object in the image. The image type of the area to which the pixel at the position belongs is detected. Therefore, the region determination can be performed accurately and easily as compared with the conventional method in which the region determination is performed using the statistical method based on the bitmap image data.

  In this embodiment, the image is divided into regions for each image type in units of lines. However, the present invention is not limited to this, and may be divided into regions for each image type in units of pixels. In this case, the process of S16 described above may be omitted. In this way, the processing amount increases because the number of area determination processes increases, but by increasing the determination frequency, the learning table 32 can be quickly changed and the compression rate can be further improved.

  In the image data encoding apparatus 1 of the present embodiment, the recording unit 31 stores three probability estimation tables 34 for the text area, the graphics area, and the bitmap area. Then, the table changing unit 23 uses the probability estimation table 34 referred to by the arithmetic encoding unit 26 when the encoding target pixel shifts to a region of a different image type, and the image type of the region including the encoding target pixel. The configuration is switched to the probability estimation table 34 according to the above.

  Since the probability estimation table 34 includes update information of the learning table 32, it can be said that an update method (learning method) of the learning table 32 is determined. Therefore, with the above configuration, the optimization method of the learning table 32 can be changed for each image type, and the learning table 32 can be optimized efficiently.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims. That is, embodiments obtained by combining technical means appropriately modified within the scope of the claims are also included in the technical scope of the present invention.

  According to the present invention, image data can be encoded at a high compression rate. Therefore, the present invention can be suitably used for various devices that need to compress and transmit image data, such as a facsimile transmitter or an output device that outputs image data to an image forming apparatus.

1, showing an embodiment of the present invention, is a functional block diagram illustrating a configuration of an image data encoding device. FIG. It is a figure explaining the processing content of arithmetic coding. FIG. 1 is a diagram illustrating an embodiment of the present invention and illustrating how an image is divided into regions for each image type. It is a figure which shows the positional relationship of the pixel of encoding object, and the surrounding pixel used for a context. It is a figure which shows the content of the learning table. It is a figure which shows the content of the probability estimation table. FIG. 4 is a flowchart illustrating an embodiment of the present invention and illustrating a first half of processing by an image data encoding device. FIG. 7 is a flowchart illustrating an embodiment of the present invention and showing a middle part of processing by an image data encoding device. FIG. 7 is a flowchart illustrating an embodiment of the present invention and illustrating a latter part of processing by an image data encoding device. The other embodiment of this invention is shown and the content of the recording part is shown. FIG. 6 shows still another embodiment of the present invention, and shows the contents of a recording unit. FIG. It is a block diagram which shows a prior art and shows the structure of QM-Coder. It is a flowchart which shows a prior art and shows the flow of a process by QM-Coder.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Image data encoding apparatus 11 Drawing command reception part (drawing command reception means)
12 Area information creation unit (area information creation means)
13 Image data creation unit (image data creation means)
21 Context generation part (extraction means)
22 Area determination unit (image type detection means)
23 Table changing section (learning table changing means, table switching means)
24 Acquisition unit (encoding means)
26 Arithmetic coding unit (encoding means, loop means)
27 register (data holding means)
28 Area information adding unit (area information adding means)
31 recording unit (learning table storage means, probability estimation table storage means, initial table storage means)
32 learning table 33 initial learning table (initial table)
34 Probability estimation table 35 Area information

Claims (15)

In an image data encoding apparatus that sequentially encodes symbols constituting image data as encoding target symbols, and sequentially encodes the encoding target symbols by arithmetic encoding based on symbol appearance probabilities estimated from surrounding symbol patterns ,
Learning table storage means for storing a learning table showing a correspondence relationship between the pattern of the peripheral symbols and an index for specifying the symbol appearance probability;
A probability estimation table storage means for storing a probability estimation table in which a correspondence relationship between the index shown in the learning table, the symbol appearance probability, and update information indicating update contents of the index in the learning table is shown;
Extraction means for extracting the encoding target symbol and the peripheral symbol pattern from the image data;
Image type detection means for detecting the image type of the partial area in the image corresponding to the encoding target symbol extracted by the extraction means;
A learning table for changing a learning table stored in the learning table storage unit when the image type detected by the image type detection unit is different from the image type previously detected by the image type detection unit Change means,
The index corresponding to the pattern of the peripheral symbols extracted by the extraction unit is acquired from the learning table, the appearance probability and the update information corresponding to the acquired index are acquired from the probability estimation table, and the acquired appearance The encoding target symbol extracted by the extraction unit based on the probability is encoded by arithmetic encoding, and the index included in the learning table is updated based on the encoding target symbol and the acquired update information. Encoding means for
An image data encoding apparatus comprising: the extraction unit; the image type detection unit; the learning table change unit; and a loop unit that causes the encoding unit to repeatedly execute processing. An initial table storage means for storing an initial table serving as a template for the learning table;
The learning table changing means is stored in the learning table storage means when the image type detected by the image type detecting means is different from the image type previously detected by the image type detecting means. The image data encoding apparatus according to claim 1, wherein the learning table is rewritten based on the initial table stored in the initial table storage unit. The initial table storage means stores a plurality of the initial tables for each image type,
The learning table changing unit encodes the learning table when the image type detected by the image type detecting unit is different from the image type previously detected by the image type detecting unit. The image data encoding apparatus according to claim 2, wherein rewriting is performed based on the initial table corresponding to the image type of the partial area corresponding to the target symbol. The learning table storage means stores the learning table for each image type,
The learning table changing unit refers to and updates by the encoding unit when the image type detected by the image type detection unit is different from the image type previously detected by the image type detection unit. The image data encoding apparatus according to claim 1, wherein the learning table is switched to a learning table corresponding to an image type of the partial area corresponding to the encoding target symbol. A drawing command receiving means for receiving a drawing command for instructing to place a desired type of object in a desired position in the image in order to create the image;
Image data creating means for creating the image data according to the drawing command received by the drawing command receiving means;
2. The image according to claim 1, wherein the image type detection unit detects an image type of the partial region corresponding to the encoding target symbol based on a drawing command received by the drawing command receiving unit. Data encoding device. From the drawing command received by the drawing command receiving means, the type of the object arranged in the image and the range occupied by the object in the image are acquired, and the image is based on the type and range of the acquired object. Is further provided with region information creating means for creating region information indicating the image type of the divided partial image and the range of the partial image in the image,
6. The image according to claim 5, wherein the image type detection unit detects an image type of the partial region corresponding to the encoding target symbol based on the region information created by the region information creation unit. Data encoding device.   The image processing apparatus further comprises region information adding means for adding the region information to encoded image data obtained by encoding all symbols constituting the image data. The image data encoding device according to claim 6. The probability estimation table storage means stores a plurality of the probability estimation tables for each image type,
When the image type detected by the image type detection unit is different from the image type previously detected by the image type detection unit, the probability estimation table referred to by the encoding unit is 2. The image data encoding apparatus according to claim 1, further comprising table switching means for switching to a probability estimation table corresponding to the image type of the partial area corresponding to the encoding target symbol. Data holding means for holding an arithmetic code obtained by sequentially encoding the encoding target symbols by the encoding means;
The encoded data holding unit outputs the arithmetic code when the image type detected by the image type detection unit is different from the image type previously detected by the image type detection unit. The image data encoding apparatus according to claim 1, wherein the arithmetic code held by itself is reset.   The image data encoding apparatus according to claim 1, wherein the encoding unit performs encoding according to a QM-Coder system.   The data output means for outputting the encoded image data obtained by encoding the image data by the encoding means to the image forming apparatus. 2. The image data encoding device according to 1. Each of the above means is realized by a computer executing a program,
6. The image data encoding apparatus according to claim 5, wherein the drawing command receiving unit receives the drawing command from an application program via an operating system.   The program for functioning a computer as each means of the image data coding apparatus of any one of Claim 1 to 12.   A computer-readable recording medium on which the program according to claim 13 is recorded. In an image data encoding method in which a symbol constituting image data is an encoding target symbol, and the encoding target symbol is sequentially encoded by arithmetic encoding based on a symbol appearance probability estimated from a pattern of surrounding symbols.
An extraction step in which the extraction means extracts the encoding target symbol and the pattern of the peripheral symbols from the image data;
An image type detection unit for detecting an image type of a partial region in the image corresponding to the encoding target symbol extracted in the extraction step;
When the image type detected in the image type detection step is different from the image type previously detected in the image type detection step, a learning table changing unit is stored in a storage unit, A learning table changing step for changing a learning table showing a correspondence relationship between a pattern of surrounding symbols and an index for specifying the symbol appearance probability;
The encoding means obtains the index corresponding to the pattern of the peripheral symbols extracted in the extraction step from the learning table stored in the storage means, and stores the learning stored in the storage means From the probability estimation table showing a correspondence relationship between the index shown in the table, the symbol appearance probability, and update information indicating update contents of the index in the learning table, the appearance probability corresponding to the acquired index and the Update information is acquired, and the encoding target symbol extracted in the extraction step based on the acquired appearance probability is encoded by arithmetic encoding, and the encoding target symbol and the acquired update information are used as described above. Coding to update the index included in the learning table And a step,
An image data encoding method comprising repeatedly executing the extraction step, the image type detection step, the learning table change step, and the encoding step.

Images