JP6965539B2 - Coding device, decoding device and program - Google Patents

Description

The present invention relates to a coding device, a decoding device and a program.

Patent Document 1 describes a program that compresses information as a result of numerical analysis. In this program, the arithmetic unit reads the values of a plurality of predetermined structural parameters used in the numerical analysis from the storage unit as variable-dependent information, and the arithmetic unit generates conversion information from the variable-dependent information. The unit includes a step of converting a plurality of result information based on the conversion information, a step of the arithmetic unit compressing the converted result information, and a step of the arithmetic unit storing the compressed result information in the storage unit. ..

Patent Document 2 describes a method of compressing the spatial coordinates of particles and their time-series data. In this method, the process of converting the coordinates from the information of the space to be analyzed and the particles existing there to the expression of the required accuracy and the process of dividing a part of the information of the converted coordinates into a group to be shared. And the process of outputting the minimum necessary information for each group.

Patent Document 3 describes an apparatus for losslessly compressing data having various distributions including data having a large variance and data having non-stationary time. This device acquires an input value for each time from input data composed of time-series input values, and calculates a predicted value for each time using a predetermined prediction method based on the past input value from the acquired input value. Then, the residual calculation unit that calculates the residual between the acquired input value and the calculated predicted value for each time by the processing device, and the coding method indicated by each of the plurality of coding parameters indicating different coding methods. A coding method in which each of a plurality of coding parameters indicates a prior probability storage unit that stores prior probability information that defines prior probabilities in advance by a storage device, and a residual that is past the residual calculated by the residual calculation unit. Based on the code length of the past residual when encoded using, and the prior probability of the coding method defined in the prior probability information stored in the prior probability storage unit, the residual calculation unit An coding method is shown in which the code length of the past residual is shorter than that of other coding methods when the past residual is encoded among a plurality of coding parameters for the calculated residual. The coding parameter selection unit that selects the coding parameters for each time by the processing device and the residuals calculated by the residual calculation unit are acquired for each time, and the acquired residuals are obtained by the coding parameter selection unit. It is encoded by the processing device using the coding method indicated by the coding parameter selected for the difference, the code word of the residual is calculated for each time, and the calculated code word is output as compressed data of the input data. It is provided with a residual coding unit.

Japanese Patent No. 4493614 Japanese Unexamined Patent Publication No. 9-160898 Japanese Patent No. 5570409

As a code suitable for data compression, a variable length code (entropy code) is known in which a symbol having a higher probability of appearance is assigned a shorter code. In the coding and decoding by the variable length code, the coding and decoding are performed according to the code table in which the symbol and the code are associated with each other. Since the code table becomes bloated as the number of symbols contained in the data increases, it becomes an obstacle to high compression efficiency.
In view of the above circumstances, an object of the present invention is to perform coding and decoding by a variable length code with higher efficiency than using a code table constructed based on an empirical distribution.

The invention according to claim 1 is an estimation amount calculation means for calculating an estimated amount of a population of a probability distribution of the input data from the input data and a distribution model of the input data, and a probability using the calculated estimated amount. By the appearance probability calculation means for calculating the appearance probability of the data value included in the input data from the distribution, the coding means for variable-length coding the data value using the calculated appearance probability, and the coding means. As an incidental code attached to the variable-length encoded data value, a coding device including profile information for specifying the distribution model and an output means for outputting the estimated amount is provided.

In the coding apparatus according to claim 1, the invention according to claim 2 converts the input data represented by a fixed-point number or a floating-point number into input data represented by an integer, and is represented by an integer. It includes the estimation amount calculation means and the integer conversion means for inputting the input data to the coding means.

In the invention according to claim 3, in the coding apparatus according to claim 1 or 2, the difference in the time direction of the input data is calculated, and the difference is used as the estimated amount calculating means and the reference numeral in place of the input data. A difference calculation means to be input to the conversion means is provided.

In the invention according to claim 4, in the coding apparatus according to any one of claims 1 to 3, the input data is divided into frames for each time step, and the input data is divided into frames for each time step. A calculation means and a frame division means to be input to the coding means are provided.

The invention according to claim 5 is the coding apparatus according to any one of claims 1 to 4, wherein the accuracy of the input data is adjusted according to a specified accuracy, and the accuracy of the input data is adjusted. Is provided with the estimator calculating means and the accuracy adjusting means for inputting to the coding means.

In the invention according to claim 6, in the coding apparatus according to any one of claims 1 to 5, the data value is divided into groups based on variable dependence information related to the data value, and each group is divided into groups. A group dividing means for inputting the input data to the estimated amount calculating means and the coding means is provided.

The invention according to claim 7 is a sampling means according to any one of claims 1 to 6, in which a sample is extracted from the input data and the extracted sample is input to the estimator calculation means. To be equipped.

According to the invention of claim 8, in the coding apparatus according to any one of claims 1 to 7, the input data is converted into upper digits and lower digits based on the number of data and significant figures of the input data. A precision dividing means for dividing and inputting the upper digit into the coding means is provided.

The invention according to claim 9 is input by an input means for inputting profile information for specifying a distribution model of input data and an incidental code indicating an estimated amount of a population of a probability distribution of the input data, and the input means. The appearance probability calculating means for calculating the appearance probability of the data value included in the input data by using the distribution model specified from the incidental code and the estimated amount to be restored, and the calculated appearance probability. It is used to provide a decoding device including a decoding means for decoding the variable length encoded data value.

The invention according to claim 10 uses a computer as an estimation amount calculating means for calculating an estimated amount of a population of a probability distribution of the input data from the input data and a distribution model of the input data, and the calculated estimated amount. An appearance probability calculating means for calculating the appearance probability of a data value included in the input data from the probability distribution used, a coding means for variable-length coding the data value using the calculated appearance probability, and the code. As ancillary code attached to the data value variable length encoded by the conversion means, profile information for specifying the distribution model and a program for functioning as an output means for outputting the estimated amount are provided.

The invention according to claim 11 uses the input means for inputting the profile information for specifying the distribution model of the input data and the incidental code indicating the estimated amount of the population of the probability distribution of the input data, and the input means. An appearance probability calculating means for calculating the appearance probability of a data value included in the input data using the distribution model specified and the estimated amount to be restored from the input incidental code, and the calculated said Provided is a program for functioning as a decoding means for decoding the variable length encoded data value by using the appearance probability.

According to the inventions according to claims 1 and 10, coding by a variable length code is performed with higher efficiency than using a code table constructed based on an empirical distribution.
According to the invention of claim 2, even if the data is represented by a fixed-point number or a floating-point number, coding with a variable-length code is more efficient than using a code table constructed based on an empirical distribution. It is done in.
According to the invention of claim 3, even if the input data does not follow a specific distribution, if the difference of the input data follows a specific distribution, the coding of the difference by the variable length code will make the empirical distribution. It is performed more efficiently than using the code table constructed on the basis.
According to the invention of claim 4, even when the distribution of the input data changes with the passage of time, the coding by the variable length code is more efficient than using the code table constructed based on the empirical distribution. It is done in.
According to the invention of claim 5, the compression ratio is improved as compared with the case where the accuracy of the input data is not adjusted.
According to the invention of claim 6, the compression rate is improved as compared with the case where the data value is not divided into groups based on the variable-dependent information.
According to the invention of claim 7, the calculation amount of the estimated amount is reduced and the memory is saved as compared with the configuration in which the estimated amount is calculated using all the data.
According to the invention of claim 8, when the empirical distribution of the input data is sparse, the amount of calculation required for the code design of the variable length code is reduced as compared with the case where all the digits are input to the coding means. NS.
According to the inventions of claims 9 and 11, decoding with a variable length code is performed with higher efficiency than using a code table constructed based on an empirical distribution.

The figure which shows the hardware configuration of the information processing apparatus 1. The figure which shows the functional structure of the information processing apparatus 1. The flow chart which shows the procedure of the process which the information processing apparatus 1 executes. A graph showing the empirical distribution of velocity at a given time step. The figure which shows the correspondence table of a profile number and a distribution model. The figure which shows the probability density function using the calculated estimator. A graph showing the probability density function. The figure which shows the structure of the coded data of a body code. The figure which shows the structure of the coded data of ancillary code. Correspondence table between profile number and distribution model. The figure which shows the probability density function using the restored estimator. A graph showing the probability density function. The figure which shows the functional structure of the modification. The figure which shows the conversion of a fixed-point number. Diagram showing the conversion of floating point numbers. An example of converting a fixed-point data value to an integer. The figure which shows the functional structure of the modification. The figure which shows the example of the difference. The figure which shows the functional structure of the modification. An example of the input data to be coded. The figure which shows the functional structure of the modification. The figure which shows the data value before and after the precision adjustment. The figure which shows the functional structure of the modification. The figure which shows the data value before and after group division. The figure which shows the state of the particle before and after group division. The figure which shows the functional structure of the modification. The figure which shows the functional structure of the modification. The figure which shows the input data.

An example of a mode for carrying out the present invention will be described.
FIG. 1 is a diagram showing a hardware configuration of the information processing device 1. The information processing device 1 includes a calculation unit 11, a storage unit 12, a communication unit 13, an operation unit 14, and a display unit 15. The storage unit 12 is a storage device such as a hard disk drive or a memory, and stores programs and data. The arithmetic unit 11 includes a processor and a memory used as a work area for arithmetic operations, and executes arithmetic operations according to a program stored in the storage unit 12. The communication unit 13 is a communication interface between the information processing device 1 and the external device. The operation unit 14 includes an input device such as a keyboard and a mouse, and accepts operations by the user. The display unit 15 includes a display device such as a liquid crystal display panel, and displays a GUI (Graphical User Interface) screen. The operation unit 14 and the display unit 15 may be configured as external devices.

FIG. 2 is a diagram showing a functional configuration of the information processing device 1. Each of the illustrated components shows the software installed in the information processing apparatus 1 as a module for each function. The functions of these modules are realized when the arithmetic unit 11 executes the software. The distribution model selection unit 21, the estimator calculation unit 22, the appearance probability calculation unit 23, and the coding unit 24 are functions related to coding and are examples of the coding device according to the present invention. The appearance probability calculation unit 41 and the decoding unit 42 are functions related to decoding, and are examples of the decoding device according to the present invention.

FIG. 3 is a flow chart showing a procedure of processing executed by the information processing apparatus 1. (A) shows the coding procedure, and (b) shows the decoding procedure.

The distribution model selection unit 21 selects a distribution model corresponding to the input profile number (step S11). The estimator calculation unit 22 (an example of the estimator calculation means) calculates an estimator of the population parameter of the probability distribution of the input data from the input data and the distribution model of the input data (step S12). The appearance probability calculation unit 23 (an example of the appearance probability calculation means of the coding device) calculates the appearance probability of the data value included in the input data from the probability distribution using the calculated estimated amount (step S13). The coding unit 24 (an example of the coding means) encodes the data value with a variable length using the calculated appearance probability (step S14).

The appearance probability calculation unit 41 (an example of the appearance probability calculation means of the decoding device) restores the distribution using the distribution model of the input data and the estimator of the parameter of the probability distribution of the input data (step S21), and the input data. The appearance probability of the data value included in is calculated (step S22). The decoding unit 42 (an example of the decoding means) decodes the variable-length encoded data value using the calculated appearance probability (step S23). The details of each part will be described below.

[Configuration related to coding]
The configuration related to coding will be described. The input data to be encoded is, for example, the velocity Vx (i, t) in the x-axis direction obtained by performing a molecular dynamics simulation of the motion of I atoms over a T-time step under a thermal equilibrium state. ) Data value. i (i = 1, ..., I) represents the identification number of the atom, and t (t = 1, ..., T) represents the time step. The data value is represented by an integer having N significant digits.

FIG. 4 is a graph showing the empirical distribution of velocity at a given time step. The horizontal axis is the velocity and the vertical axis is the probability of appearance. In molecular dynamics simulations, the velocity distribution of atoms in thermal equilibrium is controlled to be normal.

In the molecular dynamics simulation, the velocity Vy (i, t) in the y-axis direction and the velocity Vz (i, t) in the z-axis direction are also calculated. ), Only the example of Vx (i, t) will be described below.

Next, the distribution model selection unit 21 will be described.
FIG. 5 is a diagram showing a correspondence table between the profile number and the distribution model. In this example, a total of five types of distribution models, a normal distribution, an exponential distribution, a Poisson distribution, a gamma distribution, and a Weibull distribution, are shown, but other distribution models may be included in the correspondence table. The correspondence table is set in the distribution model selection unit 21.

A profile number is input to the distribution model selection unit 21 (see FIG. 2). The profile number may be input by the operation unit 14, but the profile number may be input by providing a bit indicating the profile number in the header of the input data and reading the header.

The distribution model selection unit 21 selects a distribution model corresponding to the input profile number q from the correspondence table, and outputs distribution model information indicating the selected distribution model. For example, in a molecular dynamics simulation, the velocity distribution of atoms in the thermal equilibrium state is controlled to be a normal distribution, so 0 is input as the profile number, the normal distribution is selected as the distribution model, and the distribution shows the normal distribution. Model information is output.

Next, the estimator calculation unit 22 will be described. The estimator is an estimate of the population parameter of the probability distribution. The parameter is a constant that determines the probability distribution, such as mean, variance, standard deviation, median, and expected value. In this embodiment, the expected value and the estimated variance are calculated.
The input data and the distribution model information output from the distribution model selection unit 21 are input to the estimator calculation unit 22. The estimator calculation unit 22 calculates the expected value and the estimated amount of variance by the following procedure using the input data and the distribution model information.

Next, the appearance probability calculation unit 23 will be described. Hereinafter, the random variable that the velocity Vx (i, 1) follows is defined as V. In addition, Vx (i, 1) may be represented _{by v i.} X in the above-mentioned explanation of the estimator calculation unit 22 corresponds to _{V and v i, respectively.}

The estimator calculated by the estimator calculation unit 22 is input to the appearance probability calculation unit 23. The appearance probability calculation unit 23 calculates the appearance probability of the value that the random variable V can take from the probability density function using the estimated amount.

Next, the coding unit 24 will be described. The input data and the appearance probability calculated by the appearance probability calculation unit 23 are input to the coding unit 24.

FIG. 8 is a diagram showing the structure of coded data of the main body code. FIG. 9 is a diagram showing the structure of the coded data of the incidental code. _{The coding unit 24 reversibly encodes the data value v i} using a conventional method such as a Huffman code or an arithmetic code based on the calculation result of p (s), and outputs the data value v i as a main body code. Further, the coding unit 24 encodes the profile numbers q and the estimated quantities a and b by a fixed-length coding method, and outputs them as incidental codes. The output main body code and incidental code are transmitted to and stored in the storage unit 12 of the information processing device 1, an external storage device, or the like.

[Configuration related to decryption]
Next, the configuration related to decoding will be described. An incidental code is input to the appearance probability calculation unit 41 (see FIG. 2). The appearance probability calculation unit 41 restores the profile numbers Q, the estimators A and B from the incidental codes.

The appearance probability calculation unit 41 calculates the appearance probability of the value that the data value can take from the probability density function using the restored estimator.

Next, the decoding unit 42 will be described. The main body code is input to the decoding unit 42 (see FIG. 2). Decoding unit 42, a body code was decoded using conventional techniques such as Huffman decoding or arithmetic decoding on the basis of the calculation results of P (g), and outputs the data value d _i. The decoded data value is transmitted to and stored in the storage unit 12 of the information processing device 1, an external storage device, or the like.

The above is the processing for the data value of one time step. Assuming that the distribution of random variables is the same in all time steps, the estimator calculated by the above process from the data value of one time step (for example, t = 1) is used for the other. Coding and decoding of data values in time steps may be performed.

The difference between the present invention and the prior art will be described. For example, when Huffman coding is used in the conventional method, the size C (bit) of the code table in which the symbol and the code are associated is given by the equation (9), where N is the number of effective digits of the data value. That is, as the number of significant digits increases, so does the size of the code table. On the other hand, when Huffman coding is used in the variable length coding of the present embodiment, the profile number and the estimated amount are sent instead of the code table, but the data amount of the profile number and the estimated amount is the code table. It is smaller than the amount of data in, and is constant regardless of the number of effective digits of the data value. Therefore, according to the present embodiment, the coding and decoding by the variable length code is performed with higher efficiency than using the code table constructed based on the empirical distribution.

[Modification example]
The above embodiment may be modified as follows. A plurality of modified examples may be combined.

[Modification 1]
FIG. 13 is a diagram showing a functional configuration of a modified example. This modification is an example in which the integer conversion unit 31 is added to the above embodiment. The integer conversion unit 31 (an example of an integer conversion means) converts a data value represented by a fixed-point number or a floating-point number into a data value represented by an integer. Data values represented by integers are input to the estimator calculation unit 22 and the coding unit 24.

FIG. 14 is a diagram showing conversion of fixed-point numbers. Let D be the fixed-point number, N be the number of significant digits, and I be the value obtained by converting D to an integer. At this time, I is calculated by the equation (10).

FIG. 15 is a diagram showing the conversion of floating point numbers. Assuming that the floating-point number is F, F is represented by the equation (11) by the formal value M and the exponential value E. Let M and E be the converted integer data values.

FIG. 16 is an example of converting a fixed-point number data value into an integer. (A) is a fixed-point number data value, and (b) is an integer data value.

According to this configuration, even data represented by a fixed-point number or a floating-point number can be coded by a variable-length code with higher efficiency than using a code table constructed based on an empirical distribution. For the decrypted data, the decoding unit 42 converts the data represented by an integer into the data represented by a fixed-point number or a floating-point number by the inverse conversion of the above conversion.

[Modification 2]
FIG. 17 is a diagram showing a functional configuration of a modified example. FIG. 18 is a diagram showing an example of the difference. This modification is an example in which the difference calculation unit 32 is added to the above embodiment. The difference calculation unit 32 (an example of the difference calculation means) calculates the difference in the time direction of the input data. The moving distance Δx from the x-coordinate value x (t) of the particle at the time t to the x-coordinate value x (t + Δt) after the minute time Δt is the velocity v _{x in the x-axis direction of the time as shown in the equation (12).} Is proportional to. Therefore, when _{the distribution model of v x} has a normal distribution, the distribution model of the difference (movement distance) in the time direction of the coordinate data also has a normal distribution. The difference calculated by the difference calculation unit 32 is input to the estimator calculation unit 22 and the coding unit 24.

According to this configuration, even if the input data does not follow a specific distribution, if the differences in the input data follow a specific distribution, the coding of the differences with variable length codes is constructed based on the empirical distribution. It is more efficient than using a code table.

[Modification 3]
FIG. 19 is a diagram showing a functional configuration of a modified example. This modification is an example in which the frame dividing portion 33 is added to the above embodiment.

FIG. 20 is an example of input data to be encoded. This input data is, for example, the velocity in the x-axis direction obtained by performing a molecular dynamics simulation of the motion of I (I = 50,000) atoms over a T-time step (T = 100,000) under a thermal equilibrium state. It is a data value of Vx (i, t). i (i = 1, ..., I) represents the identification number of the atom, and t (t = 1, ..., T) represents the time step. The velocity Vx (i, t) is represented by a fixed-point number with 6 significant digits from -9.99999 to 9.99999.

The frame dividing unit 33 (an example of the frame dividing means) divides the input data into frames for each time step. In the example of FIG. 20, each is divided into T frames including the data values of the velocity Vx of the atoms of i = 1, ..., I. Input data is input to the estimator calculation unit 22 and the coding unit 24 for each frame, and the processing illustrated in the above embodiment is executed for each frame.

According to this configuration, even when the distribution of the input data changes with the passage of time, the coding by the variable length code is performed with higher efficiency than using the code table constructed based on the empirical distribution.

[Modification example 4]
FIG. 21 is a diagram showing a functional configuration of a modified example. This modification is an example in which the precision adjusting unit 34 is added to the above embodiment. Here, the precision is a digit indicated by the lowest value of the mantissa part in a floating-point number. For example, if the data value is 1.2345e-05, then 1.2345e-05 = 12345e-09, so the accuracy is e-09. The user determines in advance the accuracy required for the decrypted data (hereinafter referred to as the required accuracy). The required accuracy may be input by the operation unit 14, but the required accuracy may be input by providing a bit indicating the required accuracy in the header of the input data and reading the header.

The accuracy adjusting unit 34 (an example of the accuracy adjusting means) adjusts the accuracy of the data according to the input required accuracy. Specifically, the accuracy adjustment unit 34 divides the data value into upper digits and lower digits according to the input required accuracy. The high-order digit is the part from the highest-order digit to the digit corresponding to the required precision. The lower digit is a part from the digit immediately below the digit corresponding to the required accuracy to the lowest digit.

FIG. 22 is a diagram showing data values before and after the accuracy adjustment. (A) the accuracy unadjusted data values v _i 'is, (b) is the data value v _i "after accuracy adjustment. Precision adjustment unit 34, v _i' according to the required accuracy ε formula ( discard the lower digit by 13), to calculate the v _i ". The accuracy adjustment data whose accuracy has been adjusted by the accuracy adjustment unit 34 is input to the estimator calculation unit 22 and the coding unit 24. According to this configuration, the compression ratio is improved as compared with the case where the accuracy of the input data is not adjusted.

[Modification 5]
FIG. 23 is a diagram showing a functional configuration of a modified example. This modification is an example in which the group division unit 35 is added to the above embodiment. The group dividing unit 35 (an example of the group dividing means) groups the data values based on the variable dependency information related to the data values. Variable dependency information is another variable on which one variable depends. For example, in the molecular dynamics method, the velocity distribution of a particle depends on the mass, so the mass is variable-dependent information with respect to the velocity. A group is a group of data classified by variable-dependent information. For example, the velocity distribution of ideal gas particles in a thermal equilibrium state changes depending on the mass of the particles. That is, mass is variable-dependent information for velocity.

FIG. 24 is a diagram showing data values before and after group division. FIG. 25 is a diagram showing the state of particles before and after group division. For example, FIG. 24 (A), FIG. 25 if the particles having a particle mass m _B having a mass m _A as shown in (A) are mixed, the group division unit 35, the input data has a mass m _A group of the particles is divided into groups of particles having a mass m _B (FIG. 24 (b), 25 (group a shown in b)) (FIG. 24 (c), the group B shown in FIG. 25 (c)).

Grouped data values are input to the estimator calculation unit 22 and the coding unit 24, the processing illustrated in the above embodiment is executed for each group, and the estimator and the appearance probability are calculated for each group. FIG. 25 (e) is a probability density function using the estimator of group A, and FIG. 25 (f) is a probability density function using the estimator of group B. FIG. 25 (d) is a probability density function using an estimator calculated without grouping. According to this configuration, the compression ratio is improved as compared with the case where the data values are not divided into groups based on the variable dependency information.

[Modification 6]
FIG. 26 is a diagram showing a functional configuration of a modified example. This modification is an example in which the sampling unit 36 is added to the above embodiment. The sampling unit 36 (an example of sampling means) extracts data used for calculating the estimated amount. In the above embodiment, the estimator is calculated from i = 50,000 particles, but for example, 1000 samples are extracted from 50,000 particles. The extracted samples are input to the estimator calculation unit 22, and the estimator is calculated from these samples. Estimators are not calculated for data other than the extracted samples, and coding is performed using a code table. According to this configuration, the calculation amount of the estimated amount is reduced and the memory is saved as compared with the configuration in which the estimated amount is calculated using all the data.

[Modification 7]
FIG. 27 is a diagram showing a functional configuration of a modified example. In this modification, the precision dividing unit 37 and the precision integrating unit 51 are added to the above embodiment, and the first coding unit 241 and the second coding unit 242 are provided in place of the coding unit 24, and the decoding unit 42 is provided. This is an example in which the first decoding unit 421 and the second decoding unit 422 are provided instead of the above. If the accuracy of the input data is high, if the number of valid digits of the input data is not sufficient, or if the data distribution is large, the empirical distribution of the input data becomes sparse. The efficiency of coding with respect to the computational amount of variable length coding is reduced. Therefore, in this modification, the data value is divided into upper digits and lower digits, and the target of variable length coding is limited to the upper digits. Specifically, the precision dividing unit 37 (an example of the precision dividing means) divides the input data into upper digits and lower digits based on the number of data of the input data and significant figures.

FIG. 28 is a diagram showing input data. (A) is the input data v _i "before division. (B) is the upper digit v _{high, i} " defined by the equation (14). (C) is the lower digit v _{low, i} "defined in the equation (15). The upper digit is input to the first coding unit 241 and the second coding unit 242 is filled with the upper digit. , The lower digit is input. The first coding unit 241 encodes the upper digit v _{high, i} into the first code by the variable length code. The second coding unit 242 is the lower digit. The digit v _{low, i} ”is encoded into a second code with a fixed length code. The first code is input to the first decoding unit 421, and the second code is input to the second decoding unit 422. The first decoding unit 421 restores the upper digit by decoding the first code based on the appearance probability. The second decoding unit 422 restores the lower digit by decoding the second code. The precision integration unit 51 integrates the restored upper and lower digits. According to this configuration, when the empirical distribution of the input data is sparse, the amount of calculation required for the code design of the variable length code is reduced as compared with the case where all the digits are input to the coding means.

[Modification 8]

[Modification 9]

[Modification 10]

[Modification 11]

[Modification 12]

[Modification 13]
The program for causing the computer to execute the above processing may be provided, for example, in a state of being continuously stored in a computer-readable recording medium such as an optical recording medium or a semiconductor memory, or a communication such as the Internet. It may be provided over the network. When the program according to the present invention is provided in a state of being continuously stored in a recording medium, a computer reads the program from the recording medium and uses it. When the program according to the present invention is provided via a communication network, the computer receives the program from the distribution source device and uses it.

1 ... Information processing device, 11 ... Calculation unit, 12 ... Storage unit, 13 ... Communication unit, 14 ... Operation unit, 15 ... Display unit, 21 ... Distribution model selection unit, 22 ... Estimator calculation unit, 23 ... Appearance probability calculation Unit, 24 ... coding unit, 241 ... first coding unit, 242 ... second coding unit, 31 ... integer conversion unit, 32 ... difference calculation unit, 33 ... frame division unit, 34 ... accuracy adjustment unit, 35 ... group division unit, 36 ... sampling unit, 37 ... accuracy division unit, 41 ... appearance probability calculation unit, 42 ... decoding unit, 421 ... first decoding unit, 422 ... second decoding unit, 51 ... accuracy integration unit

Claims (11)

An estimator calculating means for calculating an estimator of the population parameter of the probability distribution of the input data from the input data and the distribution model of the input data.
An appearance probability calculation means for calculating the appearance probability of a data value included in the input data from a probability distribution using the calculated estimated amount, and an appearance probability calculation means.
Encoding means for variable length coding the data value using the calculated the probability of occurrence,
As ancillary code attached to the data value variable-length coded by the coding means, a profile information for specifying the distribution model, an output means for outputting the estimated amount, and an output means for outputting the estimated amount.
A coding device comprising. Integer conversion in which the input data represented by a fixed-point number or a floating-point number is converted into input data represented by an integer, and the input data represented by an integer is input to the estimated amount calculating means and the coding means. The coding apparatus according to claim 1, further comprising means. The coding apparatus according to claim 1 or 2, further comprising a difference calculating means for calculating a difference in the time direction of the input data and inputting the difference to the estimator calculating means and the coding means in place of the input data. .. The item according to any one of claims 1 to 3, further comprising a frame dividing means for dividing the input data into frames for each time step and inputting the input data to the estimator calculating means and the coding means for each frame. The encoding device described. Any of claims 1 to 4, further comprising an accuracy adjusting means for adjusting the accuracy of the input data according to a designated accuracy and inputting the adjusted input data to the estimator calculating means and the coding means. The encoding device according to item 1. Claims 1 to 1, further comprising a group dividing means for dividing the data value into groups based on variable dependence information related to the data value and inputting the input data to the estimated amount calculating means and the coding means for each group. 5. The encoding device according to any one of 5. The coding apparatus according to any one of claims 1 to 6, further comprising a sampling means for extracting a sample from the input data and inputting the extracted sample to the estimated amount calculation means. 3. The encoding device according to any one item. An input means for inputting profile information for specifying a distribution model of input data, and an incidental code indicating an estimated amount of a population parameter of the probability distribution of the input data.
An appearance probability calculation means for calculating the appearance probability of a data value included in the input data by using the distribution model specified and the estimated estimator to be restored from the incidental code input by the input means.
Using the calculated the probability of occurrence, and decoding means for decoding the variable length encoded said data values,
Decoding device. Computer,
An estimator calculating means for calculating an estimator of the population parameter of the probability distribution of the input data from the input data and the distribution model of the input data.
An appearance probability calculation means for calculating the appearance probability of a data value included in the input data from a probability distribution using the calculated estimated amount, and an appearance probability calculation means.
A coding means for variable-length coding the data value using the calculated appearance probability, and
A program for functioning as an output means for outputting profile information for specifying the distribution model and the estimated amount as an incidental code attached to the data value variable-length coded by the coding means. Computer,
An input means for inputting profile information for specifying a distribution model of input data, and an incidental code indicating an estimated amount of a population parameter of the probability distribution of the input data.
An appearance probability calculation means for calculating the appearance probability of a data value included in the input data by using the distribution model specified and the estimated estimator to be restored from the incidental code input by the input means.
A program for functioning as a decoding means for decoding the variable-length encoded data value using the calculated appearance probability.

Images