JP2000022552A - Information processing device and system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform effective compression for coding by calculating a coding error from the code word length that is decided based on the occurrence frequency and correcting the occurrence frequency of another character string based on the calculated coding error. SOLUTION: The information on the occurrence frequency of a partial character string stored in a dictionary 102 is used to analyze the characters by a character string analyzer 101 and also to generate codes by a code generator 103. The analyzer 101 reads the candidate character having high occurrence probability after an inputted character string out of the dictionary 102 based on the transition probability that is set between the character strings stored in the dictionary 102. The generator 103 replaces the signals which are inputted by a signal input device 100 and the analyzer 101 with the code words and performs the optimum generation of codes based on the occurrence frequency of the character string stored in the dictionary 102. Then a coding error is calculated to correct the occurrence frequency of other character strings whose generation of code words are not over yet. This arithmetic processing is successively repeated for the signals having higher occurrence frequencies.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an encoding / decoding apparatus, and more particularly to a processing procedure and an apparatus configuration for efficiently compressing a signal.

[0002]

2. Description of the Related Art TA Welch describes an encoding method for character strings such as documents and programs as "A Technique for
High Performance Data Compression ”, IEEE Comput
ing, vol. 17, pp. 8-19, June, 1984. Since this is an improvement on the conventional method proposed by Lempel and Zif, the three initials are taken and LZW
It is called a method. This method is characterized in that it has a procedure of decomposing an input character string (signal sequence) into a partial character string (partial signal sequence) and registering it in a dictionary based on the frequency of occurrence. Thereby, if the input partial character string has already been registered in the dictionary, data compression is realized by converting the character string into a registration number of the dictionary. Another feature of this method is that it can adapt to the characteristics of a character string by updating the dictionary contents based on the number of occurrences of the input character string.

On the other hand, from a viewpoint different from the above, a method of generating a codeword called a Huffman code is described in Huffman, DA
But: “A method for the construction of minimumre
dundancy codes ", Proceeding of IRE, vol. 40, no.
9, pp. 1098-1101, Sep. 1952. This method is a procedure for generating a code based on a signal occurrence probability. When the number of events that can be taken by a signal increases, it takes an enormous amount of processing time for code generation. Therefore, in general, all codes are created in advance and used in a table. For example, facsimile defines all codes as international standards. This is a code created based on the Huffman method from a continuous number of black and white (run length) obtained by defining a typical image and performing raster scan.

[0004]

Generally, the compression processing comprises a modeling unit for converting an input signal sequence and a code conversion unit for allocating code words. Conventionally proposed LZW
The method is a method in which the processing of the former modeling unit is devised for character strings. Also, the conventional Huffman code is
This is a typical method of the latter code conversion unit using the occurrence probability of a signal. As described above, all of the conventional methods have a problem that sufficient compression efficiency cannot be realized because the modeling unit and the code conversion unit are separated.

Further, the above-mentioned conventional system is a system which performs an operation irrelevant to signal generation means. That is, the procedure of the compression process does not change when the signals to be encoded are all generated in advance (batch type) or when the operator inputs characters one by one (real time type).
In the case of the real-time type, the exchange of signals is not considered in the input operation of the operator and the compression processing procedure.

An object of the present invention is to provide an information processing apparatus that performs efficient compression in encoding.

[0007]

An object of the present invention is to provide a dictionary for storing at least an input unit for inputting a character, a character string, the number of occurrences of the character string, and an identification number given to each character string in association with each other. The code word length is determined based on the number of occurrences of the character string stored in the dictionary in order from the character string having the highest number of occurrences, and the code word of the corresponding identification number is generated from the code word length, and the obtained code word is determined. This can be achieved by providing a coding unit that obtains a coding error from the length and corrects the number of occurrences of another character string from the coding error.

[0008] Further, the object of the present invention is to code a character string having the largest number of occurrences from a dictionary storing an input character string, the number of occurrences of the character string, and an identification number given to each character string. Determining the bit length of the word, generating a code word of the identification number of the character string from the integer part of the bit length, and determining the encoding error from the determined bit length,
Determining a new occurrence count of the remaining character string from the encoding error and the occurrence count of the remaining character string, and determining the bit length of the code word from the occurrence count for the character string having the largest occurrence count of the remaining character string. And generating a code word of the identification number of the character string from the integer part of the bit length.

That is, the code word generated according to the present invention corrects the number of occurrences of a character string that has not been generated each time a code word is generated. By doing so, the generated codewords are close to each other, and an efficient code can be generated. In addition, a codeword can be generated at high speed because the codeword is generated from the integer part of the obtained code length.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In the following embodiments, a basic configuration of a batch type and a real time type will be described. A specific embodiment of input support in a real-time type will be described. In any case, the feature is that an optimum codeword is generated according to the property of the input signal, and a specific procedure of the code generation will be described.

In the following embodiment, an input signal is a character code series (character string) represented by one byte. However, the present invention targets signals such as characters, images, and voices,
Widely applicable.

In the following description, there is a place where the words signal, event, character, or a character string and a signal sequence are used interchangeably. In the present invention, since the occurrence probability and the number of occurrences can be mutually converted, there are some places where the same meaning is used.

(1) Basic Configuration FIG. 1 shows a basic configuration of a part for performing an encoding process of an information processing apparatus according to the present invention.

The signal input device 100 operates in a batch type or a real time type. In the case of the batch type shown in FIG. 1A, a signal sequence that is stored in a file in advance is an input target. The batch type does not require a support function for the operator. On the other hand, in the case of the real-time type as shown in FIG. 1 (2), for example, the operator inputs a signal by performing a key operation for each character using a keyboard. In the real-time type, a display unit or the like is provided to display a candidate character estimated from an input character string to the operator, thereby assisting the operator in inputting a signal.

The character string analyzer 101 reads out from the dictionary 102 a candidate character string having a high probability of occurring subsequent to an input character string, using the transition probability between characters registered in the dictionary 102. In addition, the character string input from the signal input device 100 is decomposed into partial character strings by using the transition probability between characters registered in the dictionary 102.

The dictionary 102 is a means for storing data relating to the structure of a partial character string. The transition probability between characters or the number of occurrences of a partial character string is stored using a data structure such as a tree format. The contents of the dictionary 102 can be fixedly set in advance, or can be updated using the measurement result of the number of occurrences of the partial character string decomposed by the character string analysis device 101. Code generation device 10
No. 3 converts the partial character string decomposed by the character string analysis device 101 into an efficient codeword by referring to the information on the number of occurrences registered in the dictionary 102.

In the present invention, information on the number of occurrences of a partial character string stored in the dictionary 102 is stored in the character string analyzing apparatus 10.
1 is used for character string analysis and code generation of the code generation device 103.

In real-time signal input, the registered contents of the dictionary 102 are used to support input by the operator in the signal input device 100.

According to the present invention, a high compression rate is realized by generating an optimal code based on the number of occurrences, and the size of the apparatus is reduced by using a dictionary for a plurality of purposes.

(2) Configuration of the Signal Input Device 100 The basic operation of the signal input device 100 is an input / output interface for sequentially inputting a signal such as a character string and passing the signal to the character string analyzer 101.

The real-time signal input device 100 includes:
In addition to the above-mentioned batch type operation, input means for each character, means for recognizing handwritten characters, kana-kanji conversion means, display means for input characters and input candidate characters, instruction input means for controlling the operation of the device, etc. Works in combination with.

For example, the signal input device 100 includes a keyboard and a display. When the operator inputs one character as in the procedure shown in FIG.
101 reads out a partial character string starting from the input character in the dictionary 102 as a candidate and displays it on the display of the signal input device 100. Each time a character is input, candidates for partial character strings registered in the dictionary are narrowed down. If the operator wants to enter a character string in the partial character string displayed on the display, the operator does not have to enter all the characters by keyboard by selecting the character string. If there is no candidate, the operator inputs a partial character string using the keyboard one character at a time. The character string thus selected or input is transmitted to the encoding device 103 as an encoding target.

As a result of such a procedure, it is possible to execute character input simply and at high speed without the operator having to input all the characters on the keyboard. In addition, displaying candidate characters registered in the dictionary helps to input a character string based on ambiguous memory. For example, when converting kana-kanji,
By displaying candidate characters, it is possible to prevent erroneous input due to conversion mistake of homonyms. The present invention can provide input support for such an operator.

(3) Configuration of Character String Analysis Device 101 The character string analysis device 101 decomposes a character string input from the signal input device 100 into partial character strings. Further, a candidate character string having a high probability of occurring subsequently to the input character string is read from the dictionary 102. In each case, the transition probability between characters registered in the dictionary 102 is used.

FIG. 3 shows the configuration of an input / output unit and a code generation unit of the information processing apparatus. When a real-time signal is input, as shown in FIG. A partial character string that is temporarily stored in the input character buffer 106 and registered in the dictionary 102 is searched using a dictionary search device. In the figure, the English text “I HAVE APE
N. "Is being input. In the dictionary 102,
By storing information that the first character "I" is likely to be a first-person subject represented by one character, and secondly to be a space indicating a character delimiter, the subject "I" A verb starting with "HA"
E, HAD, HATE, etc. are read from the dictionary as candidates and stored in the candidate character buffer 107. The input characters of the buffer 106 and the candidate characters of 107 are displayed on a display provided in the signal input device 100. When the operator selects one of the candidates, the input character can be determined without inputting one character at a time. Candidate character buffer 1
07 is stored in the input character buffer 106.
Go to The code generation device 103 generates a code based on the number of occurrences of the partial character string in the dictionary 102 with the input character buffer 106 as a target of the encoding process, and stores the generated code in the code memory 105 or outputs it to the outside. .

In the case of the real-time type, a character string can be decomposed using a control signal input by an operator. For example, as shown in the operation example of the signal input device 100, the terminal of the determined character can be set as a character string delimiter by an operation of selecting one of the displayed candidate character strings. In addition, a conversion confirmation input of kana-kanji conversion in Japanese input can be used as a signal for instructing a character string delimiter.

FIG. 4 shows an input unit and a code generation unit of the information processing apparatus. In both the batch type and the real-time type, as shown in FIG. The input character string can be decomposed based on the information on the transition probability between the characters of the input character string in the dictionary 102 only from the column to create a partial character string. This includes
Using the input character string, the tree is traced from the beginning, and when it reaches the end of the tree, it is determined that a partial character string is delimited. Characters such as periods, commas, spaces, and parentheses included in a character string can be used as a signal indicating a character string delimiter. In addition, character strings can be decomposed based on part-of-speech determination based on grammar, analysis of subject predicates, and the like. As the configuration of the character string decomposing device, a conventional LZW character string decomposing method can be used. The decomposed partial character string is converted into a code by the code generation device 103.

(4) Configuration of Dictionary 102 The dictionary 102 stores information on the structure of partial character strings and the number of occurrences. The dictionary 102 supports character string input for simplifying the input operation of the operator in the signal input device 100, and creates a partial character string in the character string analysis device 101,
It is used for the compression processing of the code generation device 103 and the like. When a dictionary is configured with a tree structure, a partial character string can be represented by allocating one character to one branch and tracing the branch to the end via a branch point. FIG. 5 schematically shows a tree structure made up of a partial character string constituting the character string "I HAVE A PEN." And some other partial character strings. Each branch or branch point is provided with a counter so that the number of occurrences can be written. The substring is a single character like "I",
It may be composed of a plurality of characters such as “PEN”, or may be a sentence.

According to the present invention, code generation is calculated using the number of occurrences counted by the counter. Although it can be converted to a probability value, a circuit for performing a decimal point operation is required. For example, by dividing the number of occurrences of a branch connected from a branch point to a branch (that is, a character and a character) by the total number of occurrences of the branch output from the branch point, a transition probability between characters can be obtained. . For a plurality of branches (ie, character strings) connected via a plurality of branch points, the occurrence probability of the character string can be obtained by combining the transition probabilities between the characters.

In the initial state, a uniform number of occurrences can be assigned to all partial character strings. Alternatively, the number of occurrences of a partial character string can be set in advance by statistically measuring a typical document in advance. Alternatively, the result of the previous operation can be stored and used as an initial value. If the entire signal to be encoded is stored in a file in advance, the signal is pre-scanned, the signal characteristics such as the number of occurrences of partial character strings are measured, and a dictionary is created. An encoding process can be performed based on the content.

The above-mentioned initial setting value can be fixedly used, or the content of the counter can be updated based on the number of occurrences of a newly input partial character string. For example, if the input partial character string is already registered in the dictionary 102, the occurrence count of the corresponding counter is updated. If not registered, new registration is performed by increasing the number of branches. A partial character string whose occurrence count does not increase can be deleted. For the contents of the dictionary 102, the registered contents and the update time are set so that both the encoding device and the decoding device can refer to the common dictionary contents. Although the data capacity of the dictionary 102 fluctuates due to the update of the dictionary contents, there is no problem if the data is stored in a rewritable memory. The data structure for memory storage is not limited.

When the input signal is a 1-byte character code, the maximum number of branches at each branch point of the tree structure is 256 (2
Raised to the eighth power) 65536 (2 to 1) for a 2-byte code
6). Furthermore, control codes such as start and stop can be incorporated. These control codes can be converted into codes like character codes.

For the purpose of miniaturizing the dictionary 102, for example, by limiting the partial character string to two characters, a dictionary using only the number of transitions between characters can be constructed. As shown in FIG. 6, the character set 2 is composed of a combination of characters before and after the transition.
Consists of a dimension table. Count the number of transitions,
While updating the counter in the table, the combination with a small number of appearances is deleted. The counter initial value can be set based on a fixed number or a measurement result of a typical document. When an unregistered character occurs, for example, “1” is assigned as an initial value of the number of transitions and newly registered.

The dictionary 102 can be constructed by adding information on the number of occurrences to items registered in a general dictionary such as a Japanese dictionary, an English-Japanese / Japanese-English dictionary, a personal name and place name dictionary, and the like. For example, as shown in FIG. 7, the general dictionary 401 is stored in a read-only memory, and the occurrence count dictionary 402 is stored in a rewritable memory. Then, the candidate character string is read from the general dictionary 401 using the dictionary search device 305, and information on the number of occurrences of the candidate character is read from the occurrence number dictionary 402. When the number of items that have actually occurred is small, the number of registrations in the occurrence count dictionary 402 is
The number is smaller than the number of items in the general dictionary 401. Thus, the dictionary 10 is used for both input support and character string compression.
2 can be used, and the size of the apparatus can be reduced.

Information about the tree structure, the number of occurrences, or the probability of occurrence is stored in a rewritable semiconductor memory such as an SRAM or FERAM, so that the contents of the dictionary are updated based on the result of statistically measuring the input character string. can do. The data structure for storing in memory is
For example, a conventionally known link structure or the like can be used, but is not particularly limited.

As described above, according to the present invention, the dictionary 10
2, the character input support and the encoding process can be executed, and the function of the general dictionary can be shared.

(5) Basic Configuration of Code Generation Device 103 The code generation device 103 replaces the signals input by the signal input device 100 and the character analysis device 101 with code words. The present invention performs optimal code generation using the number of occurrences of a character string in the dictionary 102. The partial character string to be encoded may be one character, word, sentence, etc., and each is converted into one codeword and output. The timing of the encoding process can be executed in real time type each time a partial character string is input, or can be executed in batch type for the entire character string stored in a file in advance.

The present invention is characterized in that a code is generated by a binary operation, and a decimal point operation based on a probability value is not required, so that signal processing can be performed at high speed with a simple device configuration.

First, referring to FIG. 8, a binary code (s = 2)
The code generation procedure in the case of is described. The numerical values used in the description are examples, and the actual values may vary depending on conditions.

First, (a) Ci (i = 1 to n) is obtained by arranging the number of occurrences of n types of character strings in descending order. The number of characters constituting each character string is not limited. As an example, the total number A of the number of occurrences C is set to 1024 which is a power of two.

A = ΣCi = 1024 (1) The character strings to be processed are in descending order of the number of occurrences, and i = 1 is set as an initial value.

(B) Bit length bi of code word of character string i
Is set by the following equation (base is 2 and the decimal part is rounded off).

Bi = log _s (1 / pi) = log _s (A / Ci) = log _s (A) −log _s (Ci) = 10−log ₂ (Ci) (2) It is noted that it is not necessary to calculate the logarithmic operation log ₂ (Ci) of the above equation to the decimal point, since an integer value is obtained. For example, input Ci
Means for inputting the numerical value of and outputting the logarithm calculation result calculated in advance.

However, when 2 ^bi obtained in the case of rounding is replaced with the actual number of occurrences of the event, if there is a change in the rank of the number of occurrences of the event, 1 is added to bi ( Goin). In other words, even if the number of occurrences of the event is replaced, the binary number having the minimum number of digits that does not change in the order is obtained, and the number of digits bi is used as the number of bits of the code word.

When converting a probability value given as a decimal number to an integer number of bits (code length), a reduction in the compression ratio occurs due to round-up processing. Here, the difference between the probability value pi and the minimum information amount of the set code word is obtained by the following equation. Ei = (1 / s) ^bi-pi (3) The encoding error is calculated using the number of occurrences Ci. It is calculated by the following equation.

CEi = A · Ei = A · ((1 / s) ^bi−pi ) = 2 ^{(10−bi) −Ci} (4) (c) For the purpose of keeping the value of the cumulative occurrence count constant Then, the generated encoding error is subtracted from the number of occurrences of the remaining character strings (from i + 1 to n). Here, the method of distributing the encoding error is not limited, but is conditioned on the fact that the order of the number of occurrences of the event of the distribution destination does not change. For example, distribution is performed in proportion to the ratio of the number of occurrences of the remaining character strings.

(D) The procedure from (b) is repeated using the corrected number of occurrences until the code length of the actually generated character string to be encoded is generated. When the processing of the event that has actually occurred is completed, the code length bi is set, and the process proceeds to (e).

(E) When the code length of the character string to be coded is determined, code bits are set. First, the cumulative occurrence count is calculated by the following equation. Then, 10 bits (this value is log _s (A) = 10 which is a setting example of equation (1)) are expressed in binary notation, and the upper empty bits are set to 0. For example, 512 in decimal notation is "100 000 000 000". And
The number of bits bi is cut out from the highest order, and a code word is set.

Qi = Σ (2 ^bk ) (sum of k = 1 to i−1) (5): End The reason why the above procedure is optimal is that, at first, in each stage of the code generation procedure, it is based on the occurrence probability. The shortest codeword length is set. By allocating the error of the probability value generated by setting the codeword length to the probability values of the remaining events, no coding loss occurs. By repeating this procedure sequentially, an optimum codeword can be generated.

FIG. 9 shows the configuration of the code generation device 103. A signal to be encoded is input from the signal input device 100, and the number of occurrences is input from the dictionary 102. The generated code is output to a code memory or a transmission device. Here, the code length setting unit 110 is means for realizing the setting of the code length in the above equation (2). The input to the code length setting unit 110 is a signal Ci (i = 1 to n) representing the number of occurrences,
In the above example, the minimum is 0 and the maximum is 1024. The values to be output are the minimum 0 and maximum 10 values indicating the code length. As an example of an apparatus configuration for performing this processing, the input number of occurrences is temporarily stored in a register, and
The codeword length is set by reading a memory in which the operation results are summarized in a conversion table. Next, the sign bit setting unit 111
A code word that can be uniquely decoded is set by setting the cumulative number of occurrences represented by the above equation (5) in a register and cutting out the contents of the register with the set code word length. The codeword length conversion unit 112 calculates (1/3) of the first term on the right side of the above equation (3).
s) means for calculating ^bi , which temporarily stores the code word length input from the code length setting unit 110 in a register, and calculates a value corresponding to the exponentiation operation by a binary number generating circuit such as a binary number shift or bit setting. .

The number-of-occurrence correcting section 113 temporarily stores the number of occurrences of a character string, calculates the encoding error of the above equation (4), and generates the remaining character strings for which codeword generation has not been completed. This is a means for correcting the number of times. In these devices, the management device 115 performs the operation in order from the signal with the largest number of occurrences, based on the code generation procedure.
Processing is terminated when a code corresponding to the input signal is generated, and the code is output. The management device 115 also controls input / output such as initialization, signal input, and code output, and controls processing timing between the devices.

As shown in FIG. 10, the occurrence count table is constituted by a rewritable memory, and the occurrence count read from the dictionary 102 is set as an initial value. Each time one code length is calculated according to the code generation procedure, the occurrence count table is rewritten. Therefore, a correction value is generated by a correction value calculation circuit, temporarily stored in a correction value register, and the occurrence count table is rewritten based on the contents of the register.

FIG. 11 shows a configuration example of the correction value calculation circuit. First, the above (3) generated by the codeword length conversion unit 112
Enter (1 / s) ^bi of the first term on the right side of the equation and the number of occurrences. Then, the coding error of the above equation (4) is calculated by comparing with the content of the occurrence count register. Next, the distribution error of the encoding error is calculated based on the distribution ratio proportional to the number of occurrences. Here, the distribution error is (1 / s) ^bi
It is the following binary number. Then, a correction value is calculated by adding / subtracting the distribution error to / from the content of the occurrence count table. These processes are controlled by the management device 115.

A feature of the present invention is that, in the above code generation procedure and apparatus configuration, all operations are handled by integer values (binary numbers). A signal processing procedure or a processing device that handles a decimal point requires, for example, a floating-point arithmetic circuit. On the other hand, the present invention requires only binary integer arithmetic, so that the device configuration is easy and high-speed processing is possible. . Also, for example, the calculation of the codeword length based on the equation (2) is set in a conversion table as shown in FIG. 12A, and stored in a memory as shown in FIG. Is read out from the memory as an address to obtain the code length.

It is needless to say that the operation contents of these device configurations can also be realized by signal processing by a processor under program control.

Next, an example of generating a binary code using actual numerical values will be described with reference to FIG. In the figure, code generation examples of registration numbers 1 and 2 are shown for eight types of partial character strings. As an initial setting, the number of occurrences of each character string is set to 1, and thereafter, a count value of the number of occurrences of the input character string is set. The total number of occurrences is 1024, which is a power of two.

First, (a) the number of occurrences Ci of n = 8 types of character strings is changed to 500, 200, 100, 100, 50, and 5 in the order of size.
0, 23, and 1. The last character string (occurrence number Ci =
1) shows a case where the initially set value remains, and is an example in which the value is not actually input. The total number A is 1024.

(B) The number of occurrences of a character string of i = 1 is 500
From Equation (1a), the code length is bi = 1.
Even if 2 ^bi = 512 is replaced with the original number of occurrences, there is no rank change. From the equation (2a), the coding error Cei = 12.

(C) The encoding error is converted to the remaining character string (i = 2
From 8 to 8) and correct. Specifically, the number of occurrences of the character string i = 1 is determined by the coding error Cei = 12.
Therefore, 12 is subtracted from the number of occurrences of the remaining character strings. Since the distribution method is not limited as described above, an example is shown.

(D) If the actually input character string is not i = 1, the process returns to step (b) to process the next character string. On the other hand, when the input character string is i = 1, the code length b1 = 1 is set, and the procedure proceeds to the step (e).

(E) When the character string actually input is i = 1, the code length is b1 = 1 and the cumulative number of occurrences is q1 = 0. When the cumulative number of occurrences is represented by a 10-bit binary number, "00000
00000 ", and code length b1 = 1 bit is taken out from the upper bit, and" 0 "is a code word. When the character string is i = 2, the code length is b2 = 2 and the cumulative occurrence count is q2 = 512. When the cumulative occurrence count is represented by a 10-bit binary number, "10
00000000 ", and takes out 2 bits from the upper
"0" is a code word.

End: (6) Decoding Process FIG. 14 shows an example of the configuration of an encoding unit that performs the encoding process of the processing device. The decoding device is a character string analysis device 1
1, a dictionary 102, a code generation device 103, and a match determination device 104. Here, the character string analysis device 101, the dictionary 102, and the code generation device 103 have the same configuration as the coding unit. The codewords stored in the code memory 105 are read, and it is sequentially determined whether or not the codewords generated in the order of the registration numbers are the same. If they are the same, a character string corresponding to the registration number is output. Then, the number of occurrences of the character string is calculated by measuring the decoded signal, and the same tree structure as that of the encoding device is created and stored in the dictionary. The signal can be reversibly reproduced by referring to this dictionary. Here, the same contents as those of the encoding process are executed at the same timing for the initial setting and the like.

FIG. 15 shows the decoding procedure. Each step is the same as the content of the encoding process, and includes a step of determining whether or not the last input code word is the same as the code word generated in the order of the registration number.

A feature of the present invention is that, in the above procedure and apparatus configuration, all operations are handled by integer values (binary numbers) as in the case of the encoding processing. Compared with a signal processing procedure or a processing device that handles a decimal point, the configuration is easy and high-speed processing is possible.

Next, for one input code word,
A basic numerical example of the decoding process is shown. It is assumed that the number of occurrences is set in advance.

The beginning (a) From the measurement result of the already decoded character string, the number of occurrences Ci of n = 8 types of character strings is 500,
200, 100, 100, 50, 50, 23, and 1. The total number A is 1024.

(B) The number of occurrences of a character string with i = 1 is 500
From Equation (1a), the code length is bi = 1.
Even if 2 ^bi = 512 is replaced with the original number of occurrences, there is no order change. From the equation (2a), the coding error Cei = 12.

(C) Encoding error is converted to the remaining character string (i = 2
From 8 to 8) and correct. More specifically, since the number of occurrences of the character string i = 1 is set to be larger by 12, the number of occurrences of the remaining character string is 1
Subtract 2 An example is shown because the distribution method is not limited.

(D) Since the code length is bi = 1 and the cumulative occurrence count is q1 = 0, "0" is the code word.

(E) Both the input code sequence and the generated code word are compared over the length of the number of bits bk. If they match, the code word indicates the character string i = 1, and the decoding process ends. If they do not match, the procedure from (b) is repeated for the next character string using the corrected occurrence count.

End: By replacing the number of occurrences of a signal having a certain number of occurrences or less, or a signal having a certain number of ranks or less with a uniform numerical value, the generation of code bits can be simplified. For example, if the character string is 25
When there are six types (2 to the eighth power), the upper eight types of character strings use the measured number of occurrences, and the remaining 24
A uniform number of occurrences is assigned to eight types of events.
However, the total number of occurrences must be maintained. A character string with a uniform number of occurrences (in the example above,
(Up to 56th), it is possible to perform encoding and decoding processing at high speed by utilizing the fact that a uniform codeword length can be allocated.

(7) Example of Device Configuration An example of a device configuration using the present invention will be described.

FIG. 16 shows the configuration of a system including a plurality of information processing apparatuses. This system shows a configuration example in which efficient data transmission is performed between a transmitting device and a receiving device connected by a transmission path. In the dictionaries provided in both the transmitting device and the receiving device, an initialization condition and an updating method are set in common for both. This can be set based on mutual communication prior to code transmission, or can be set based on an initialization program incorporated in the device in advance.

The dictionary contents can be used fixedly or updated based on the signal contents. When updating the dictionary contents, the transmitting side updates the dictionary based on the coded and transmitted signal. On the other hand, the receiving side updates the dictionary based on the received and decoded signal. In this way, the updated content of both dictionaries is always matched with the timing, so that the code transmitted by the transmitting device can be decoded without error by the receiving device to reproduce the signal.

The speed at which the operator inputs characters using, for example, a keyboard is slower than the data transfer speed of the transmission path. In such a case, occupying the transmission path for transmitting the codeword reduces the efficiency of use of the transmission path. Therefore, packet-type communication known as one of communication procedures can be used as a method of transmitting codes. In this method, the amount of data to be transmitted (for example, the number of bytes) is charged, and is not related to a connection time such as a telephone. Therefore, it is suitable for sending a signal input by the operator in real time in a fragmented manner based on the input timing. According to the present invention, the transmission cost can be reduced by transmitting a code that is input and compressed in real time by packet communication.

FIG. 17 shows an example of the configuration of an information processing apparatus having a built-in encoding unit and decoding unit. By compressing the signal,
The feature of this device is that it can store a large amount of data. The code memory 105 is a means for storing code words, and can use a semiconductor memory, a disk device, or the like. The operations of the encoding process and the decoding process are as described above.

Here, all means for constituting the encoding processing unit and the decoding processing unit can be prepared separately, but a part of them can be shared. By performing timing control so that encoding and decoding do not operate at the same time, the character string analysis device 101, the dictionary 102, and the code generation device 103 can be shared for both encoding and decoding. That is, by not simultaneously executing input and output of signals, the circuit scale of the information processing apparatus can be reduced.

(8) Configuration Example of Portable Information Terminal FIG. 18A shows an example in which the information processing apparatus is applied to a portable information terminal. The signal input device 100 includes a display device 300, a pen-type input device 303, and a character recognition device 304 that converts characters input by handwriting into character codes. The contents of the dictionary 102 are referred to by the character string analyzer 101, and the candidate character strings are read and displayed on the display device 300. The dictionary contents are also referred to by the character recognition device 304 and can be used to improve the recognition rate in character code conversion of characters input by handwriting. Furthermore, the character recognition device 304 uses the handwritten characters that are difficult to distinguish in a process of converting the characters into printed characters.
By converting the data into compressed data in real time in step 3, a large amount of data can be efficiently stored in the code memory 105. As described above, in the present invention, the contents of the dictionary 102 are used for the encoding process of the input character string together with the support of the character string input.

The contents of the dictionary 102 can be updated based on the contents of the input character string, or the default dictionary contents can be fixedly used.

As shown in FIG. 18B, a character string 30 input by the operator using the pen-type input means 303 character by character.
From 1, a candidate character string 3 based on a dictionary prepared in advance
02 is taken out and displayed on the display device 300. The operator selects one of the displayed candidate characters 302. If there is no corresponding character string in the displayed candidates, the operator inputs one character at a time using the pen-type input device 303.
In the figure, as an example of English sentence input, the intention is to input the verb "HAVE" following the subject "I", and enter "H" in the middle.
A shows the operation at the time when "A" is input. From dictionary 102,
This is an example of extracting a first-person singular verb candidate character string starting from “HA” and displaying “HAVE”, “HAD”, and “HATE”. Thus, the operator can easily and accurately input by selecting candidate characters without inputting all character strings. A similar procedure can be performed when the signal input device is a keyboard. Then, at the timing when the input character string is determined, it is converted into compressed data and stored in the memory.

When the handwritten character recognition device is used, the recognition accuracy can also be improved by using the candidate character strings extracted from the dictionary 102 in the character recognition device 304 that converts handwritten characters into character codes.

An example of a circuit configuration for realizing the above configuration is shown in FIG.
Shown in In addition to a CPU, a read-only non-volatile memory, a rewritable memory, and an external input / output device, the tablet, the display device, and the input / output control device are combined as means for inputting handwritten characters. A program describing the operation of the CPU, a dictionary, a transmission control protocol, and the like are written in the read-only nonvolatile memory. On the other hand, rewritable memories are SRAM, FER
It is composed of AM or the like, and stores data in the middle of calculation, the number of occurrences of a character string based on a measurement result, an input character string, and the like. By superimposing the tablet and the display device so that the pixel positions are synchronized, the input and the display position match. In such a configuration of the portable information terminal, since the character input means of the operator is limited to the tablet, real-time input support and compression of input data, which are features of the present invention, can be efficiently realized. For example, a large amount of data can be stored by inputting characters such as a personal name, an address, and a telephone number while performing the input support, performing compression processing, and storing the data in a memory. Further, in the case of the batch type, input support is unnecessary because document data and the like are input collectively, and compression processing is performed while character string analysis is performed, and storage is performed in a memory.

For example, in the kana-kanji conversion, by presenting a plurality of kanji candidates of the same homonym, the conversion operation based on the ambiguous memory of the input operator is supported. Further, by storing or transmitting the input character string as a code word, it is possible to reduce the storage capacity or the transmission time. In the case of transmission using communication means, the receiving device can reproduce exactly the same character string as the transmitting device by sequentially decoding codewords. Thus, the dictionary 102
, Data compression can be performed together with input support by the operator.

(9) Processing for Editing Compressed Data The operation for editing such as rewriting, moving, and deleting compressed data stored in the memory will be described. In the present invention, since the generated codeword is of variable length, and the dictionary contents referred to in the encoding process are sequentially updated based on the signal input, the compressed data must be decoded sequentially from the beginning to reproduce the signal. Can not. Therefore, basically, it is not possible to extract a part of the compressed data and decode a partial character string. Some methods are described below with reference to FIG.

The first method uses a device configuration to which an editing memory is added. The input character string is stored in a code memory by performing the above-described real-time type encoding processing. In order to edit the contents of the code memory again, the code word is decoded, written into the edit memory, and the character string stored in the edit memory is edited. The edited character string is subjected to a batch type encoding process and stored in a code memory. In this method, in order to perform an editing process, decoding and encoding of the entire code data are repeated by batch-type signal processing. In the second method, as shown in the content of the code memory 105 in the figure, a delimiter signal indicating a delimiter of code data is inserted during the compression process. In this way, the compressed data is decoded for each delimiter signal, stored in the editing memory, and edited. As a result, it is possible to edit only the partial character string including the edit target without decoding the entire character string. The edited character string is compressed again and stored in the code memory.
Here, note that the capacity of the compressed data changes due to the editing process, and the compressed data is stored in the memory based on the data management means. The timing of performing the above-described delimiter processing may be a method of setting for each of a fixed number of characters, a method of setting based on paragraphs / page breaks in a sentence, and the like. The delimiter signal inserted into the compressed data can be easily detected by using a unique bit pattern.

[0086]

According to the present invention, it is possible to convert a character string to be sequentially input into an optimum code, realize efficient compression, and realize code generation at high speed with a simple apparatus configuration.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a basic configuration of a portion that performs an encoding process of a processing device according to the present invention.

FIG. 2 is a diagram illustrating a procedure of a signal input device.

FIG. 3 is a diagram illustrating a configuration of an input / output unit and a code generation unit of the information processing apparatus.

FIG. 4 is a diagram illustrating an input unit and a code generation unit of the information processing apparatus.

FIG. 5 is a diagram showing a configuration of a two-character transition dictionary.

FIG. 6 is a diagram showing a configuration of a dictionary.

FIG. 7 is a diagram showing a processing procedure of character input.

FIG. 8 is a diagram showing a code generation procedure.

FIG. 9 is a diagram illustrating a configuration of an encoding device.

FIG. 10 is a diagram showing a circuit configuration of an occurrence number correction unit.

FIG. 11 is a diagram showing a configuration of a correction value calculation circuit.

FIG. 12 is a diagram showing a configuration of a code length conversion table.

FIG. 13 is a diagram illustrating a numerical example of code generation.

FIG. 14 is a diagram illustrating a configuration example of a decoding unit that performs decoding of the information processing device.

FIG. 15 is a diagram showing a decoding procedure.

FIG. 16 is a diagram illustrating a configuration of an information processing system.

FIG. 17 is a diagram illustrating a configuration of an information processing apparatus.

FIG. 18 is a diagram showing a configuration when applied to a portable information terminal.

FIG. 19 is a diagram showing a circuit configuration of a portable information terminal.

FIG. 20 is a diagram showing a configuration when applied to a compressed data editing device.

[Description of Signs] 100: Signal input device, 101: Character string analysis device, 10
2 dictionary, 103 code generation device, 104 match determination device, 105 code memory, 106 input character buffer,
107: candidate character buffer; 110: code length setting device
111: code bit setting unit; 112: code word length conversion unit
113: occurrence frequency correction unit; 115: management device.

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Yasutaka Toyoda 7-1-1, Omika-cho, Hitachi City, Ibaraki Prefecture F-term in Hitachi Research Laboratory, Hitachi Ltd. 5J064 AA02 BB05 BC01 BC28 BD01

Claims (6)

[Claims] An input unit for inputting at least a character, a dictionary for storing a character string in association with the number of occurrences of the character string and an identification number given to each character string, and a character stored in the dictionary The code word length is calculated based on the number of occurrences in order from the character string having the highest number of occurrences for
A coding unit that generates a code word of a corresponding identification number from the code word length, obtains a coding error from the obtained code word length, and corrects the number of occurrences of another character string from the coding error. Information processing device. 2. The information processing apparatus according to claim 1, wherein the encoding unit generates a codeword from an integer part of a code length obtained from the number of occurrences. 3. The apparatus according to claim 1, further comprising: a decomposing unit that divides characters continuously input from the input unit into character strings based on a predetermined rule, wherein the dictionary is decomposed by the decomposing unit. String,
An information processing apparatus that stores the number of occurrences of this character string and an identification number given to each character string in association with each other. 4. The input unit according to claim 1, wherein the input unit receives handwritten characters, the display unit displays the characters input to the input unit, and the input unit and the display unit are integrated. Information processing device. 5. An information processing system to which a plurality of information processing apparatuses are connected, wherein at least one of the information processing apparatuses includes an input unit into which a character is input, a character string, a number of occurrences of the character string, and a character string. A dictionary for storing identification numbers assigned to respective dictionaries in association with each other; a display unit for displaying character strings stored in the dictionary; An encoding unit that generates a codeword of a corresponding identification number based on the number of times, and a transmission unit that transmits the generated codeword to the outside, at least one information processing device transmits the code from the outside A receiving unit for receiving, a dictionary for storing a character string in association with the number of occurrences of the character string and an identification number given to each character string, and decoding a code input to the receiving unit to generate an identification number Decryption unit The information processing system having an extraction unit for extracting a character string corresponding from the dictionary based on the identification number that is generated. 6. A character string having the largest number of occurrences from a dictionary storing an input character string, the number of occurrences of the character string, and an identification number given to each character string. Calculating a code word of the identification number of the character string from the integer portion of the bit length; and obtaining a coding error from the obtained bit length, and calculating the coding error and the number of occurrences of the remaining character string. Determining the number of occurrences of the remaining character string from the above, and determining the bit length of the code word from the number of occurrences for the character string having the largest number of occurrences of the remaining character string; Generating a code word of the identification number of the information processing apparatus.