JPH10126276A - Data-coding method for multi-value information source, data coder, data-decoding method for multi-value information source and data decoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To attain coding and decoding efficiency similar to that by an arithmetic coding system and to improve considerably coding and decoding speed. SOLUTION: The method is made up of a prediction-setting process, in which multi-value information source is divided into bit planes or level planes, and either of '0' or '1' of each plane is used for a superior symbol, n-sets of consecutive symbols are predicted and the value (n) is set as a predicted bit number run, a prediction result output process where either '0' or '1' signal is outputted as a prediction hit signal when the prediction of a noted series is hit, a succeeding bit string is coded, and the other signal is outputted as a blank signal when the prediction is not hit. A similar prediction-setting process and a similar prediction output process are recursively repeated by using a new reduced prediction bit number smaller than the value (n), when the prediction is not hit for a prescribed number of times. When the prediction bit, a similar prediction setting process and a similar prediction output process are repeated by using a new increased prediction bit number. Furthermore, in the case of decoding, similar algorithm is applied to decoding.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data encoding method and a data encoding device for compressing a multi-valued information source, and a data decoding method and a data decoding device for expanding data of a compressed multi-valued information source. . More specifically, the present invention relates to a method and apparatus for converting a multi-valued information source into a binary bit string and compressing the binary bit string.

[0002]

2. Description of the Related Art Conventionally, in the world of information theory dealing with binary signals consisting of "0" and "1", an arithmetic coding system is known. This arithmetic coding method is an entropy coding method and has essentially the property of lossless coding (lossless). The principle then reorganizes the ideal coding scheme for memoryless sources known as Elias coding into a practical form. That is, the arithmetic code means that a corresponding section on a straight line of “0” and “1” is divided into unequal lengths according to the occurrence probability of each symbol, and a target symbol sequence is assigned to a corresponding partial section, and recursion is performed. The coordinates of the points included in the section obtained by repeatedly performing the division are represented by at least binary decimal numbers that can be distinguished from other sections, and are used as codes as they are.

[0003] This arithmetic coding method requires less hardware such as the scale of an encoder, for example, a required memory amount, and is higher than a block code in which a specific codeword corresponds to a finite number of information source symbols. There are advantages such as efficiency can be expected and adaptive coding is easy. For this reason, in the world of information theory dealing with binary signals, this arithmetic coding method is said to be able to compress to the level closest to the entropy of the information, and is said to be the most efficient coding method. In addition, this arithmetic coding method, in particular,
It is suitable for encoding Markov information sources.

As the arithmetic coding method, a Q coder, an arithmetic code type MEL code, a Mini-Max coder, and the like have been proposed. A system called a QM coder is known as an improved version of these arithmetic codes. This QM coder complies with the Color Still Image Coding Standard (JP
EG) and the Binary Image Coding Standard (JBIG). Note that this QM coder is a code for a binary information source, and when encoding a multi-valued information source such as JPEG, preprocessing for binarizing the multi-valued information source is required. In such a case, although the number of binary symbols to be encoded increases, it is possible to convert the binary symbols into a binary sequence without increasing the information amount as a multi-level information source.

[0005] This QM coder uses JPEG and JBI.
Although the mechanism is described in detail in the definition of G, the outline thereof will be briefly described here with reference to FIG. 18 for comparison with the present invention described later. Since the configuration of the arithmetic decoding type entropy decoder is substantially the same as the configuration of the entropy encoder, the description thereof is omitted here.

A QM coder 101 serving as an arithmetic code type entropy encoder includes an arithmetic operation unit 102 and an occurrence probability generation unit 103 functioning as a state storage. In the occurrence probability generation means 103, a state parameter table necessary for determining a symbol occurrence probability required for encoding is written. The above state parameters are specified by the input state signal 106. Then, the arithmetic operation unit 102 outputs, as a read address, the data at the time of updating the operation parameter to the state parameter table specified by the state signal 106, and the occurrence probability generation unit 10
3 is output to the arithmetic operation unit 102. The arithmetic operation unit 102 compresses the input data 104 into encoded data 105 based on the data input in this way, encodes the encoded data 105, and outputs the encoded data 105. The state signal 106 is input to the occurrence probability generation unit 103. This is, for example, reference pixel data obtained by a method called a Markov model or the like, and is a signal used to increase the compression ratio.

The operation of the thus configured QM coder will be described with reference to the flowchart of FIG. First, 0xFFFF is stored in the register A in the QM coder 101.
Is substituted into the register C by 0x0000. Further, an index ST for probability estimation is initialized (step S100). Next, the symbol to be encoded (1 bit)
(Step S101). Then, it is determined whether the fetched symbol is the superior symbol or the inferior symbol (step S102). Step S when the symbol is superior
The process proceeds to step S103, and if the symbol is the inferior symbol, the process proceeds to step S106.

The probability of occurrence of the inferior symbol is obtained by referring to the probability estimation table LSZ by the index ST, and the occurrence probability of the inferior symbol is obtained by subtracting it from the register A, and the value is substituted into the register A (step S103). Thereafter, it is checked whether or not the most significant bit of the register A is "1" (step S104).
If "1", the process proceeds to step S105, and if "0", the process proceeds to step S114. When the value is “1”, the probability estimation table NMPS is referred to by the index ST,
An index ST for encoding the next symbol is obtained (step S105).

If it is determined in step S102 that the symbol is an inferior symbol, the probability of occurrence of the inferior symbol is obtained by referring to the probability estimation table LSZ using the index ST, and is substituted into the register A (step S106). Thereafter, the value of the register A is added to the register C (step S
107). Then, the probability estimation table SWITCH is referred to by the index ST (step S108),
If this is "1", the flow advances to step S109 to change the dominant symbol.

On the other hand, in step S110, an index ST for coding the next symbol is obtained by referring to the probability estimation table NLPS using the index ST. Then, in step S111, both the registers A and C are shifted left by one bit. Due to this left shift, the most significant bit overflowing from the register C is output as a code word (step S112). Then, step S1
At 13, it is checked whether or not the most significant bit of the register A is "1". If it is "1", the process returns to step S111 to repeat the shift. If the most significant bit is "0", the process proceeds to step S114, and the process ends if the encoded symbol is the last symbol. Otherwise, the process returns to step S101.

As described above, the QM coder 101 compresses and encodes a binary bit string input using the probability estimation tables LSZ, NMPS, and NLPS.

[0012]

However, this Q
The arithmetic coding method of the M coder 101 and the like has a high coding efficiency, but has a low coding speed because the coding is performed one bit at a time as shown in the flowchart of FIG. For this reason, in the practical aspect, the Lenbel-Jib coding system (= LZ coding) is dominant. However, the coding efficiency of the LZ system is considerably lower than that of the arithmetic coding system. As described above, there is no prior art in which the coding efficiency is as good as the arithmetic coding method and the code acceleration is as fast as the LZ system. Further, many coding methods such as image coding and universal coding are targeted for multi-valued information sources, and efficient compression and decompression of multi-valued information sources are demanded.

The present invention provides a new multi-valued information source that achieves substantially the same encoding and decoding efficiency as an arithmetic encoding system, greatly improves the encoding and decoding speed, and approaches the LZ system speed. It is an object of the present invention to provide a data encoding method and a data encoding device, and a data decoding method and a data decoding device for a multilevel information source.

[0014]

In order to achieve the above object, a multi-valued information source data encoding method according to claim 1 divides a multi-valued information source consisting of a plurality of bits into bit planes, and When inputting a binary bit string consisting of “0” and “1” of a plane, one of “0” and “1” is set as a superior symbol, and the other is set as an inferior symbol, and the superior symbol is used. Is n
A prediction setting step of predicting the number of consecutive bits, and setting the number n as the number of bits to be predicted; One of the signals is output as a signal per prediction, and the operation shifts to a task of encoding the next n bit strings. When the signal deviates, the other signal of either "0" or "1" is used as a codeword and an unpredicted signal. And the same prediction setting step and prediction output step are recursively repeated by setting the number of prediction bits to a new reduced prediction bit number smaller than n when the prediction is deviated a predetermined number of times. ing.

As described above, the multilevel information source is divided into bit planes, and the dominant symbol of each bit plane is n.
Are predicted to be consecutive, and when the prediction is successful, n
The bits are displayed as one signal per prediction or the like, which increases the compression efficiency and the encoding speed. In addition, when the prediction is incorrect, the number of predicted bits is reduced and the next prediction is performed. Therefore, even if the prediction is lost, the compression efficiency and the encoding speed do not decrease so much.

According to the second aspect of the present invention, the first aspect of the present invention is provided.
In the data coding method of the multi-valued information source described above, n is set to an even number, the target sequence is divided into two when prediction is deviated a predetermined number of times, and only the first half target sequence of the divided first half is inferior to the target sequence. When a symbol is present, the new reduced predicted bit number is set to の of the predicted bit number, and when the inferior symbol is present in the latter half target sequence of the second half, the new reduced predicted bit number is set to one of the predicted bit number. / 4.

In general, if there is a less-significant symbol in the second-half attention sequence, there are many cases where more less-significant symbols exist in the next attention sequence. It is possible to respond to the situation and increase the encoding speed and efficiency.

According to a third aspect of the present invention, in the data encoding method for a multi-valued information source according to the first or second aspect, when the new reduced prediction bit number is 1 and the bit is a less significant symbol, , The conventional inferior symbol is set as the superior symbol, and the conventional superior symbol is encoded as the inferior symbol. As a result, appropriate prediction can be performed according to the actual state of the input data, and high coding speed and high efficiency can be maintained.

In addition, in the invention according to claim 4, in the data encoding method for a multilevel information source according to claim 1, 2, or 3, the predetermined number of times is one. For this reason, the number of prediction bits can be reduced earlier, and the encoding speed of a bit string having a certain tendency can be increased.

According to a fifth aspect of the present invention, in the data encoding method for a multi-valued information source according to the first, second, third or fourth aspect, when the number of predictions reaches a predetermined number, the number of predicted bits is n. The number of predicted new increase bits is larger. For this reason, the more the prediction is successful, the higher the compression efficiency and the higher the encoding speed.

In the data encoding method for a multi-valued information source according to the present invention, the multi-valued information source comprising a plurality of bits is divided into bit planes, and the bit planes are divided into "0" and "1". When inputting a binary bit string
A prediction setting step of setting one of “0” and “1” as a superior symbol and the other as an inferior symbol, predicting that the n superior symbols are continuous, and setting the n as the number of prediction bits. When a prediction is hit for a sequence of interest consisting of the number of input prediction bits, either a signal of “0” or “1” is output as a signal per prediction as a signal per prediction, and the next n bit strings are output. The operation shifts to encoding, and a prediction result output step of outputting the other signal of either “0” or “1” as a codeword when it is out of order as a prediction error signal. The same prediction setting step and prediction result output step are repeatedly performed with the number of bits being the new increase prediction bit number larger than n.

As a result, when the prediction is successful, n bits are displayed as one signal per prediction or the like, and the encoding is performed for each bit plane, so that the compression efficiency is increased and the encoding speed is increased. Is faster. The more the prediction is successful, the higher the data compression efficiency and the higher the encoding speed.

According to a seventh aspect of the present invention, in the data encoding method for a multi-valued information source according to the fifth or sixth aspect, the prescribed number of times is set to two and the number of new increase prediction bits is twice the number of prediction bits. And For this reason, since the number of prediction bits is changed after the prediction hit continues, that is, after a certain tendency starts to appear, the data compression efficiency can be improved. Moreover, since the value is twice that of the previous model,
The number of bits of the consecutive dominant symbol is exactly the same due to the hit, and the probability of winning the next prediction is high. As a result, the compression efficiency can be increased and the encoding speed can be increased.

In addition, according to the invention described in claim 8, in the data encoding method for a multi-valued information source according to claim 1, 2, 3, 4, 5, 6, or 7, n is set to 2 ^m (m is (An integer of 0 or more). For this reason, when the prediction is incorrect, the division can be equally divided, and the division can be performed rapidly, and finally the number of bits can be one.

According to the ninth aspect of the present invention, there is provided the eighth aspect of the present invention.
In the data coding method of the multi-valued information source described above, the sequence of interest is divided into two when the prediction is incorrect, and “0” is set as a codeword when all of the two divided first half target sequences are dominant symbols. A first half hitting step of outputting and further dividing the latter half notice sequence of the latter half into two, and outputting a codeword of "0" or "1"; and a case where the inferior symbol exists in the first half notice sequence. Outputs a code word "1", further divides the first half target sequence into two, and outputs a code word "0" or "1". As long as there are inferior symbols, the division of the sequence of interest is repeated, and the first half hitting step and the first half off step are recursively repeated.

As a result, the same compression efficiency as that of the QM coder can be achieved, and the coding speed several times higher than that of the QM coder can be achieved.

According to a tenth aspect of the present invention, in the data encoding method for a multi-valued information source according to the eighth aspect, when the prediction is incorrect, the sequence of interest is divided into two, and the latter half of the latter half divided into two. When all the noticed sequences are dominant symbols, "0" is output as a code word, the former half noticed sequence of the first half is further divided into two, and the latter half of outputting a code word of "0" or "1" When the inferior symbol exists in the second-stage attention sequence and the second-half attention sequence, "1" is output as a codeword, and the second-half attention sequence is further divided into two to obtain "0".
Alternatively, a second-half offset process for outputting a code word of “1” is performed, and as long as the inferior symbol exists in each of the split target sequences, the split of the target sequence is repeated. Is recursively repeated.

As a result, the same compression efficiency as that of the QM coder can be achieved, and the coding speed several times higher than that of the QM coder can be achieved.

In the data encoding method for a multi-valued information source according to the eleventh aspect, the multi-valued information source comprising a plurality of bits is divided into bit planes, and the bit plane of the most significant bit is divided into the bit planes. The data is encoded by the data encoding method described in any one of the above aspects, and when "1" appears, the subsequent lower bits are output as code bits.
Therefore, in the case of an input symbol whose probability is concentrated on the lower bits, the code speed can be further increased and the required memory and the like can be reduced in size.

According to the twelfth aspect of the present invention, a multi-valued information source consisting of a plurality of bits is divided into bit planes.
In a data encoding device of a multilevel information source for compressing and encoding a binary input bit sequence consisting of "0" and "1" of each bit plane, either "0" or "1" is used. A predictive bit length and other calculation setting unit for predicting that one of the other symbols is the dominant symbol and that the other is the inferior symbol and that the dominant symbol is n consecutive, and sets the n as the number of predicted bits, and temporarily stores the input bit sequence Each value of the buffer register, the operation setting section for predicting bit length and the like, and the buffer register is input, and when a prediction is hit for a noticed sequence composed of the input number of predicted bits, “0” or “1” is used as a codeword. Either signal is output as a signal per prediction, and when it deviates, the other signal, either “0” or “1”, is output as a code word when it deviates. A prediction unit that sets a new reduced prediction bit number smaller than n by a prediction bit length etc. operation setting unit when the prediction is deviated a predetermined number of times, and sets a prediction bit when the prediction has reached a specified number of times. The number of new increase prediction bits whose number is greater than n is set by a prediction bit length etc. operation setting unit.

As a result, if the prediction is successful, n bits will be displayed as one signal per prediction, etc. In addition, since encoding is performed for each bit plane, the compression efficiency is increased and the encoding speed is increased. Is faster. Furthermore, the more the prediction is successful, the greater the number of prediction bits, so that the compression efficiency is further improved. In addition, if the prediction is incorrect, the number of predicted bits is reduced, so that even if the prediction is incorrect, the compression efficiency and the encoding speed do not decrease so much.

Further, in the data encoding apparatus for a multi-valued information source according to the thirteenth aspect, the multi-valued information source comprising a plurality of bits is divided into bit planes, and a register for temporarily storing an input bit sequence of each bit plane. A buffer register comprising a group, a register for holding a variable ofs indicating a head position of a target sequence to be encoded, and a register for holding a variable for indicating a predicted bit length, and a prediction bit length calculation unit And the position of ofs on the buffer register output from the
An input bit sequence on the buffer register determined by dth is selected, and when all the input bit sequences are predominant symbols, one predominant symbol is output as a code bit output, and when a predominant symbol is included, one inferior symbol is output. A determination unit, an input / output control signal generation unit that issues an input bit request when receiving a completion signal indicating that encoding of the input bit sequence in the buffer register has been completed from the prediction bit length calculation unit, and a value of the width that changes every moment A stack memory serving as a memory for storing the previous encoding state from the prediction bit length calculation unit, and a prediction bit for setting a prediction bit length and a dominant symbol of an input bit sequence to be newly input A length / dominant symbol setting unit. The prediction bit length and arithmetic operation symbol setting unit relative to buffer registers instructs the input bit sequence of incorporation of the number represented by the newly set new prediction bit length.

As described above, since the prediction bit length is changed by the ever-changing width, if the prediction is successful, n bits will be displayed as one signal per prediction and the like. , The compression efficiency is increased and the encoding speed is increased. Furthermore, the more the prediction is successful, the greater the number of prediction bits, so that the compression efficiency is further improved.
In addition, if the prediction is incorrect, the number of predicted bits is reduced, so that even if the prediction is incorrect, the compression efficiency and the encoding speed do not decrease so much.

In the invention according to claim 14, a multi-valued information source consisting of a plurality of bits is divided into bit planes,
In a data decoding method for inputting coded data of each bit plane and decoding it into a binary bit string consisting of "0" and "1", and then decoding a multi-valued information source, Either “0” or “1” is set as the dominant symbol, and the other is set as the inferior symbol, and the prediction result of predicting that n dominant symbols (n is an integer of 1 or more) is “0”. And an input step of inputting, one bit at a time, a codeword represented by a binary bit string consisting of "1" and "1". When the input codeword is a value per prediction, decoding n consecutive dominant symbols And when the number of prediction hits is a predetermined number of times, n
It is newly predicted that more than the dominant symbols continue.

As a result, when prediction is successful, one codeword can decode n dominant symbols. Moreover,
Since encoding is performed for each bit plane, the decompression efficiency is increased and the decoding speed is increased. Furthermore, as the prediction is more successful, the number of superior symbols that can be decoded by one codeword can be increased, so that the decompression efficiency is further increased and the decoding speed is increased.

In the data decoding method for a multi-valued information source according to the present invention, the multi-valued information source composed of a plurality of bits is divided into bit planes, and each bit plane is set to "0" or "1". One of the symbols is the dominant symbol, and the other is the inferior symbol. The prediction result of predicting that the number of the dominant symbols is n (n is an integer of 1 or more) is composed of “0” and “1”. An input step of inputting a code word represented by a bit string of a value one bit at a time, and when the input code word is a value per prediction, while decoding n dominant symbols continuously, A prediction result decoding step of inputting the next codeword when the value of the prediction error is obtained, and decoding of nm consecutive dominant symbols when the value of the next codeword is a value per prediction, and a prediction error value Again when Repeating the steps of inputting a code word recursively, so that to decode the inferior symbol when prediction off when the 0 <n-m ≦ 1.

For this reason, when the prediction is correct, n dominant symbols can be decoded by one code word, and the encoding is performed for each bit plane, so that the decompression efficiency is increased and the decoding speed is increased. . Furthermore, as the prediction is more successful, the number of superior symbols that can be decoded with one codeword can be increased, so that the decompression efficiency is further increased and the decoding speed is increased. In addition, since the operation of reducing the prediction bit length is recursively repeated at the time of misprediction, and finally the inferior symbol is decoded, data can be efficiently decoded even if misprediction occurs.

In the invention according to claim 16, a multi-valued information source consisting of a plurality of bits is divided into bit planes,
A multi-valued information for decoding a multi-valued information source after inputting a code bit as coded data of each bit plane, converting the code bit into a binary bit string consisting of "0" and "1" In the source data decoding apparatus, when one of “0” and “1” of each bit plane is set as a dominant symbol and the other is set as a dominant symbol, the dominant symbol of the code bit and the prediction bit length n are determined. A prediction bit length setting operation unit to be set; and a decoded bit for receiving a decoded bit output permission signal from the prediction bit length etc. setting operation unit, temporarily holding a code bit being input in a predetermined form, and outputting a decoded bit. When the input code bit is a superior symbol, a decoded bit output enable signal is output, the superior symbol is written in the decoded bit setting unit, and the superior symbol is output. There is when a predetermined number of times in succession have changed the prediction bit length of the number larger than n pieces.

As a result, when the prediction is correct, one codeword can decode n dominant symbols. Moreover,
Since encoding is performed for each bit plane, the decompression efficiency is increased and the decoding speed is increased. Further, as the prediction is more successful, the number of dominant symbols that can be decoded by one codeword can be increased, so that the decompression efficiency is further increased, and
The decoding speed increases.

Further, according to the seventeenth aspect of the present invention, a multi-valued information source composed of a plurality of bits is divided into bit planes, and either "0" or "1" of each bit plane is set as a superior symbol, One of the other symbols is the inferior symbol, and the number of the symbols is n (n is an integer of 1 or more)
"0" and "1" represent the results of the prediction that they are predicted to be continuous.
In a decoding apparatus for inputting, one bit at a time, code bits represented by a binary bit string consisting of: a prediction bit length and a superior symbol setting unit for setting a prediction bit length of a code bit and a superior symbol; A prediction bit length calculation unit that inputs a prediction bit length, a superior symbol, and a code bit from the length / dominance symbol setting unit, and outputs a decoded bit output permission signal according to the value of the code bit;
A decoding bit setting unit for receiving a decoded bit output enable signal, temporarily holding the input code bits in a predetermined form, and outputting decoded bits; , The dominant symbol is n
When the input code word is mispredicted, the next code bit is input. When the value is a value per prediction, the dominant symbol is n-
m pieces (m is an integer of 1 or more and smaller than n) are continuously written to the decoding bit setting section, and when the prediction is incorrect, the next code is input to the prediction bit length calculating section again.

For this reason, when the prediction is correct, n dominant symbols can be decoded with one code word, and since the encoding is performed for each bit plate, the decompression efficiency is high and the decoding speed is high. Become. Furthermore, as the prediction is more successful, the number of dominant symbols that can be decoded with one codeword can be increased, so that the decompression efficiency is further increased, and
The decoding speed increases. In addition, since the operation of reducing the prediction bit length is recursively repeated at the time of misprediction, and finally the inferior symbol is decoded, data can be efficiently decoded even if misprediction occurs.

In the data encoding method for a multi-valued information source according to the eighteenth aspect, the multi-valued information source comprising a plurality of bits is divided into level planes, and "0" of each level plane is divided.
When inputting a binary bit string consisting of
A prediction setting step of setting one of “0” and “1” as a dominant symbol, setting the other as a dominant symbol, predicting that the dominant symbol is continuous n, and setting the n as the number of bits to be predicted And, when prediction is performed on a target sequence consisting of the number of input prediction bits, one of the signals “0” or “1” is output as a signal per prediction as a codeword, and the next n bit strings are output. The encoding operation is started, and a prediction result output step of outputting the other signal of either “0” or “1” as a code word as a de-prediction signal when the prediction is deviated, and a prediction bit when the prediction is deviated a predetermined number of times. The same prediction setting step and prediction output step are recursively repeated by setting the number as the new reduced prediction bit number smaller than n.

As described above, it is predicted that n dominant symbols are continuous, and when the prediction is successful, n bits are displayed as one signal per prediction or the like, and are divided into level planes. Encoding
The encoding speed increases as the compression efficiency increases. Further, if the prediction is incorrect, the number of predicted bits is reduced and the next prediction is performed. Therefore, even if the prediction is lost, the compression efficiency and the encoding speed do not decrease so much.

In the data encoding method for a multi-valued information source according to the nineteenth aspect, the multi-valued information source composed of a plurality of bits is divided into level planes, and "0" of each level plane is divided.
When inputting a binary bit string consisting of
A prediction setting step of setting one of “0” and “1” as a superior symbol and the other as an inferior symbol, predicting that the n superior symbols are continuous, and setting the n as the number of prediction bits. When a prediction is hit for a sequence of interest consisting of the number of input prediction bits, either a signal of “0” or “1” is output as a signal per prediction as a signal per prediction, and the next n bit strings are output. The operation shifts to encoding, and a prediction result output step of outputting the other signal of either “0” or “1” as a codeword when it is out of order as a prediction error signal. The same prediction setting step and prediction result output step are repeatedly performed with the number of bits being the new increase prediction bit number larger than n.

As a result, when the prediction is successful, n bits are displayed as one signal per prediction or the like, and the encoding is performed for each level plane, so that the compression efficiency is increased and the encoding speed is increased. Is faster. The more the prediction is successful, the higher the data compression efficiency and the higher the encoding speed.

Further, in the data encoding method for a multi-level information source according to the twentieth aspect, the multi-level information source comprising a plurality of bits is divided into level planes. The planes are handled independently, and the planes with low probabilities are divided into groups in which they are grouped as a plurality of level planes. The determination bit for determining whether or not the input multi-valued information source corresponds to each group number is “0” and When inputting this judgment bit string, the input step of inputting from the group number side where the probability is concentrated and one of the judgment bits “0” or “1” of each group are input. Prediction setting in which a dominant symbol is set as one and the other is set as a non-dominant symbol, and that n dominant symbols are predicted to be continuous, and that n is set as a prediction bit number. Then, when prediction is performed on a target sequence consisting of the number of input prediction bits, one of the signals “0” or “1” is output as a codeword per prediction, and the next n bit strings are output. A prediction result output step of outputting the other signal of either “0” or “1” as a code word as a prediction error signal when the prediction is deviated, and a prediction result when the prediction is deviated a predetermined number of times. A recursive step of recursively repeating the same prediction setting step and prediction output step with the number of bits being a new reduced prediction bit number less than n; If the encoding is completed, and if not, the prediction setting step, the prediction result output step, and the recursive step It is to be executed.

As described above, only the level planes which are divided into level planes and have a high probability concentration are handled independently, and those having a low probability concentration are handled collectively by a plurality of planes. The number of planes can be reduced, and the memory for managing the predicted bit number run can be reduced. Moreover, the coding efficiency does not decrease so much. Further, it is predicted that n dominant symbols are consecutive, and when the prediction is successful, n bits are displayed as one signal per prediction or the like, so that the compression efficiency increases and the encoding speed increases. Further, if the prediction is incorrect, the number of predicted bits is reduced and the next prediction is performed. Therefore, even if the prediction is lost, the compression efficiency and the encoding speed do not decrease so much.

In addition, in the invention according to claim 21, in the data encoding method for a multilevel information source according to claim 18, 19 or 20, n is 2 ^m (m is an integer of 0 or more). . For this reason, when the prediction is incorrect, the division can be equally divided, and the division can be performed rapidly, and finally the number of bits can be one.

According to a twenty-second aspect of the present invention, in the data encoding method for a multi-valued information source according to the twenty-first aspect, when the prediction is incorrect, the sequence of interest is divided into two parts, and the first half part of the first half part is divided. When all the sequences of interest are predominant symbols, "0" is output as a codeword, and the latter half of the half-part sequence of interest is further divided into two, and the first half of outputting a codeword of "0" or "1". When the inferior symbol exists in the first half attention sequence, the codeword outputs “1” and further divides the first half attention sequence into two,
The first half has a step of outputting a code word of “0” or “1”, and as long as the inferior symbol exists in each divided sequence of interest, the division of the sequence of interest is repeated. The part removal process is recursively repeated.

As a result, the same compression efficiency as that of the QM coder can be achieved, and the coding speed several times higher than that of the QM coder can be achieved.

Further, in the data encoding apparatus for a multi-valued information source according to the present invention, the multi-valued information source comprising a plurality of bits is divided into level planes. The planes are handled independently, and the planes with low probabilities are divided into groups in which they are grouped as a plurality of level planes. The determination bit for determining whether or not the input multi-valued information source corresponds to each group number is “0” and It is composed of bits consisting of “1”, and one of the determination bits “0” or “1” is set as the dominant symbol, the other is set as the inferior symbol, and it is predicted that the number of the dominant symbols is n consecutive. A prediction bit length calculation setting section for setting the n bits as the prediction bit number, a buffer register for temporarily storing an input bit sequence, a prediction bit length calculation setting section, and the like. And the values of the buffer register are input, and when a prediction is hit for the sequence of interest consisting of the input number of prediction bits, one of the signals "0" or "1" is output as a codeword as a signal per prediction. And when it deviates, the code word is "0" or "1".
And a decision unit for outputting one of the other signals as a misprediction signal, and when the prediction is deviated a predetermined number of times, the number of predicted bits is set to a new reduced prediction bit number smaller than n by a prediction bit length etc. calculation setting unit. When the number of predictions reaches a predetermined number, the number of prediction bits is set to a new increase prediction bit number larger than n by a prediction bit length or the like calculation setting unit.

As a result, if the prediction is successful, n bits are displayed as one signal per prediction or the like, and the encoding is performed for each level / brain, so that the compression efficiency is increased and the encoding speed is increased. Be faster. Moreover, the more the prediction is successful, the greater the number of predicted bits, so that the compression efficiency is further improved. In addition, if the prediction is incorrect, the number of predicted bits is reduced, so that even if the prediction is incorrect, the compression efficiency and the encoding speed do not decrease so much. In addition, since a plurality of level planes having a low probability concentration are handled collectively, the number of level planes to be encoded can be reduced while suppressing a decrease in efficiency.

Further, in the data encoding apparatus for a multi-valued information source according to claim 24, the multi-valued information source composed of a plurality of bits is divided into level planes. The planes are treated independently, and the planes with low probabilities are divided into groups in which they are grouped as a plurality of level planes. The determination bit for determining whether or not the input multi-valued information source corresponds to each group number is “0” and A buffer register consisting of a group of registers for temporarily storing an input bit sequence consisting of this determination bit, a register for holding a variable ofs indicating the head position of the target sequence to be encoded, and a prediction bit length And a buffer register output by the predicted bit length operation unit. Position and of the ofs on other
An input bit sequence on the buffer register determined by width from s is selected, and when all the input bit sequences are predominant symbols, one predominant symbol is output as a code bit output, and when a predominant symbol is included, one inferior symbol is output. A determining unit for outputting, an input / output control signal generating unit for issuing an input bit request when receiving a completion signal indicating that encoding of the input bit sequence in the buffer register is completed from the predicted bit length calculating unit, and a constantly changing width A stack memory serving as a memory for holding the value of
Input the past encoding status from the prediction bit length calculation unit,
A predicted bit length of a newly input bit sequence and a predicted bit length / dominant symbol setting unit for setting a superior symbol, wherein the input / output control signal generation unit receives a completion signal from the prediction bit length calculation unit, The register is instructed to take in the number of input bit sequences indicated by the new prediction bit length newly set by the prediction bit length / operation symbol setting unit.

As described above, since the prediction bit length is changed by the ever-changing width, if prediction is successful, n bits will be displayed as one signal per prediction, and moreover, each level plane , The compression efficiency is increased and the encoding speed is increased. Furthermore, the more the prediction is successful, the greater the number of prediction bits, so that the compression efficiency is further improved.
In addition, if the prediction is incorrect, the number of predicted bits is reduced, so that even if the prediction is incorrect, the compression efficiency and the encoding speed do not decrease so much. In addition, since a plurality of level planes having a low probability concentration are handled collectively, the number of level planes to be encoded can be reduced while suppressing a decrease in efficiency.

In the data encoding method and data encoding apparatus for a multi-valued information source according to the present invention, the multi-valued information source is divided into bit planes or indicated by level planes divided into groups, so that a binary value is represented. Form a bit string.
Then, when the binary bit string is input, “0” or “1” is determined as a dominant symbol, and it is predicted that n dominant symbols will continue. When this prediction is successful, either “0” or “1” is output as a codeword,
Complete the encoding. If it deviates, either "0" or "1" is output, the target sequence is divided, and the signal state of each divided sequence is confirmed and encoded in the same manner as described above. . Then, the same division and prediction are repeatedly coded until prediction is achieved or division becomes impossible. In addition, the initial number of prediction bits n is changed in accordance with the hit or miss of the prediction to follow the probability fluctuation of the sequence of interest.

Further, in the data decoding method and data decoding apparatus for a multi-valued information source according to the present invention, an inverse algorithm is used for the data coding method and data encoding apparatus for a multi-valued information source described above. And decrypt it.

[0057]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to FIGS. First, compression and decompression of a binary bit string after the multi-valued information source is binarized will be described, and then the multi-valued information source will be converted into a binary bit string suitable for this compression and decompression method. The method and the like will be described.

First, the outline of the basic algorithm used in the present invention and its operation will be described with reference to FIGS. The algorithm of the present invention, like the QM coder, compresses a binary bit string. First, as an initial value, either “0” or “1” is determined as a dominant symbol, and the number ru that the symbol is predicted to be continuous is determined.
Set n. If the appearance probability of the input sequence is unknown, r
It is better to set un to 1. Then, the encoding is performed according to the following rules. Note that the number run corresponds to the number of predicted bits.

As shown in FIG. 1, all the sequences of interest indicated by "run" are predicted to be dominant symbols, and when the prediction is successful, "0" is output as a codeword and the coding of this sequence is completed. . If it deviates, "1" is output, and the next division encoding step is executed. If the prediction is incorrect, as shown in FIG.
When the first half is all dominant symbols, "0" is output as a codeword, and the coding of the first half sequence is completed. When the inferior symbol exists in the first half sequence, "1" is output as a codeword, and the next subdivision process is executed. When encoding of the first half sequence is completed, the sequence of interest is moved to the second half, and encoding is performed in the same manner as the first half sequence. For the sequence in which the inferior symbol exists, the sequence is divided as much as possible, and the above-described division encoding process is repeated.

Note that the division need not necessarily be two equal divisions, but may be non-uniform divisions or three or more divisions. In addition, when a prediction is successful, a superior symbol is output instead of “0”. When the prediction is incorrect, an inferior symbol is output instead of “1”. "May be output.

The above is the basic algorithm of data encoding of a binary bit string which is the basis of the present invention.
Further, the following processing may be added in order to follow the change in the appearance probability of the input sequence and improve the coding efficiency. That is, when the sequence predicted by run hits a predetermined number of times consecutively, for example, twice, run is increased twice or the like. If the prediction continues to be correct, the prediction range may be further expanded. Further, when the inferior symbol exists in the latter half of the sequence predicted by run, run may be reduced to 1/4 or the like.
This is because, when the inferior symbol exists in the latter half, it is determined that the following sequence includes many inferior symbols. Therefore, when the inferior symbol exists only in the first half sequence of the sequence predicted by run, a value larger than when the inferior symbol exists in the second half, for example, run
May be multiplied by １ ／. And run is 1
If it is a lesser symbol, the subsequent input sequence is inverted. That is, the dominant symbol is changed.

The encoding process of the binarized bit string used in the present invention comprises an encoding main routine shown in FIG. 4 and an encoding subroutine shown in FIG. FIG.
The encoding subroutine in the figure performs recursive reading of a function that calls the same subroutine from the subroutine.

First, each step of the encoding main routine of FIG. 4 will be described. The encoding target is an input sequence composed of binary bit strings. First, a prediction initial value run is set and a dominant symbol is selected ("0" or "1") (step S0). Next, 0 is substituted for the local variable ofs and run is substituted for the width (step S1). Here, ofs is a pointer of the array A defined in advance for encoding, and indicates a prediction start bit position. Therefore, the initial value is 0. width is ofs
Is a value indicating how many bits are to be predicted from the bit positions indicated by. In this case, the initial value of prediction run is substituted. Then, from A [ofs] of the predefined sequence A to A
Input bits are written by [width-1] (step S2). Then, from A [ofs] to A [width
-1] when all elements are dominant symbols in step S
The process proceeds to step S4, and if at least one inferior symbol is included, the process proceeds to step S5.

If the prediction is correct, a signal "0" is output per prediction as a code word, and the encoding of the sequence taken into array A is completed (step S4). On the other hand, if the prediction is incorrect, a mispredicted signal "1" is output as a code word (step S5). Then, it is detected whether the width is 1 or more (step S6). If the width is equal to or less than 1, the image cannot be further divided, so that the process proceeds to step S8 without going to the encoding subroutine of step S7. on the other hand,
If the width exceeds 1, the encoding subroutine of FIG. 5 is called (step S7).

In step S8, the prediction run is reset and, if necessary, the dominant symbol is changed. That is, in this step S8, basically, if the prediction is correct, the run is increased, and if it is off, the run is reduced. If the prediction continues to shift a predetermined number of times even when run is reduced, the dominant symbol is changed. It should be noted that various methods can be adopted as to how to evaluate a hit or a wrong prediction. For example, it is possible to adopt a method in which the run is reduced immediately when the prediction is incorrect, or the run is reduced for the first time when the prediction is incorrectly performed two or more times. Furthermore, a method in which the degree of run reduction differs between when the first half sequence or the second half sequence is deviated and when both are deviated may be adopted. It is also possible to adopt a method of drawing a predetermined probability table with an encoded bit sequence and setting the next prediction run.

If the primary prediction deviates in the encoding main routine, the encoding subroutine shown in FIG. 5 is called in step S7. The argument passed to the encoding subroutine is o
fs and width. Hereinafter, each step of the encoding subroutine will be described.

In the encoding subroutine, the prediction is divided into the first half sequence and the second half sequence, so that the prediction range is halved (step S10). That is, the width received as an argument from the parent routine is halved. Then, in the next step S11, the first half series (array A [o
fs] to A [ofs + width-1]) are all checked symbols. If all the symbols are superior, the process proceeds to step S12. If at least one inferior symbol exists, the process immediately proceeds to step S14.

If all the first half sequences are dominant symbols, "0" is output as a codeword (step S12). Then, the pointer ofs indicating the head position of the first half series is set to wi.
dth is added to change to indicate the head position of the latter half series. Further, when the first half sequence is all dominant symbols, since the inferior symbol always exists in the second half sequence, there is no need to output the code word “1” indicating that the prediction of the second half sequence has been lost. Therefore, step S2 described later
0 is skipped, and the process proceeds to step S21.

On the other hand, when the inferior symbol exists in the first half sequence, "1" is output as the code word (step S1).
4). Next, it is checked whether or not the width exceeds 1 (step S15). If the value is equal to or less than 1, the image cannot be further divided, so that the call of the child encoding subroutine (step S16) is skipped, and the process proceeds to step S17. If the width is 2 or more, it is necessary to further divide the sequence into two and encode each of them. The child encoding subroutine for that purpose is called (step S16). The child encoding subroutine is exactly the same as the encoding subroutine shown in FIG. That is, here, the same routine (function) is recursively called.

When the encoding of the first half sequence is completed by recursive calling of the encoding subroutine, the width set in step S10 is added to the pointer ofs indicating the head position of the first half sequence, and the start position of the second half sequence is indicated. It is changed (step S17). Thereafter, it is checked whether or not all the latter half sequences (from A [ofs] to A [ofs + width-1] in the arrangement) are all the dominant symbols (step S18). If all symbols are superior, step S19
Proceed to. If at least one inferior symbol exists, the process immediately proceeds to step S20. If all of the latter half sequences are dominant symbols, "0" is output as a codeword (step S19).

On the other hand, when the inferior symbol exists in the first half sequence, "1" is output as the code word (step S2).
0). Then, it is checked whether the width exceeds 1 (step S21). If less than 1,
Since the image cannot be further divided, the step S22 of executing the child encoding subroutine is skipped, and the process returns to the encoding process of the next sequence of interest. If the width of the latter half sequence is 2 or more, the sequence is further divided into two and encoded. Therefore, an encoding subroutine of the same child as the encoding subroutine shown in FIG. 5 is called (step S22). By the recursive call of this encoding subroutine, encoding of the latter half part series is executed.

A specific example of the above-described encoding process of a binary bit string will be described below. That is, as a specific example of encoding, a case will be described where the initial value of prediction run is 8, the dominant symbol is "0", and the input bit represented as "000000101" is encoded.

First, in step S2 of the encoding main routine of FIG.

The above input bits are input from [0] to A [7]. In step S3, A

[0] to A
It is determined whether all of [7] are “0”. In the case of the above example, since “1” is included in the bit string, step S
Then, the process proceeds to step 5, where "1" is output as a codeword. Subsequently, in step S6, the size of the width is checked. Since the width is 8 at this time, the process proceeds to the encoding subroutine (step S7).

In the encoding subroutine, first, in step S
At 10, the width is set to 1/2, ie, 4. Then, in step S11, the first half of the input bit, that is, A

[0]
To A [3] are all 0. In this case, since all are "0", the process proceeds to step S12, and "0" is output as a codeword. Thus, the encoding of the first half sequence is completed. Subsequently, step S13 is executed, and the process shifts to the encoding of the latter half part sequence.
Obviously, the latter half series includes "1". Therefore, unless the width is not less than 1 in step S21, the latter half sequence must be further divided and encoded. Therefore, the encoding subroutine is called again in step S22 as a child process. As a pre-process for that purpose, in step S13, the width is added to ofs, and ofs is set to the head position of the latter half series, as described above.

At step S22, ofs and width
Is called as an argument, and the child encoding subroutine is called. In step S22 of executing the child encoding subroutine,
First, step S10 of the encoding subroutine shown in FIG.
The width is further halved and changed to 2. In the next step S11, the first half sequence, that is, A [4] and A
It is checked whether [5] is both "0". In this case, since A [4] is "1", the process proceeds to the next step S14, where "1" is output as a codeword. Then, in step S15, it is determined that the width exceeds 1, and the grandchild process is called in step S16. In the grandchild encoding subroutine, first, the width is set to 1 in step S10. Since A [4] is "1", the process shifts from step S11 to step S14 to output the code word "1". In step S15, since the width is 1 or less,
Step S16 is skipped.
To 5. Since A [5] is "0", step S1
From 8 the process moves to a step S19, where the codeword "0" is output.

Next, the process exits from the grandchild encoding subroutine and returns to step S17 of the child encoding subroutine. Ofs of child encoding subroutine is 4, width
Is 2, so ofs is changed to 6 in step S17. Therefore, in step S18, A [6] and A
[7] will be checked. In this case, A [7]
Is "1", the process moves to step S20, and the code word "1"
Is output. Then, the grandchild encoding subroutine is called again in step S22. In the grandchild encoding subroutine, since A [6] is "0", the code word "0" is output in step S12. Since the width is 1, the process skips step S22 and returns to the child encoding subroutine.

The process that has returned to the child encoding subroutine further returns to the encoding main routine, and resets the predicted run and the dominant symbol in step S8. In this example, the primary prediction was missed, but the first half was hit by the secondary prediction, so run was changed from 8 to 4,
The process of making the superior symbol continue to be “0” is performed. Note that the setting of the predicted run may be changed, for example, when the prediction run is missed twice.

According to the encoding process of the binary bit string as described above, the input bits “00001001”
Is the encoded sequence of “1011010”. Therefore, in this case, the 8-bit input sequence is compressed to 7 bits.

Next, a data encoding apparatus for performing the above-described encoding process of a binary bit string will be described with reference to FIG.

This binary bit string data encoding device temporarily stores an input bit sequence 1 corresponding to the input bits described in FIGS. 4 and 5 in accordance with an instruction from an input / output control signal generation unit 5 described later. A buffer register 2 which is a register group, a determination unit 3 which inputs and compares data of the buffer register 2 and data of a prediction bit length calculation unit 4 described later, and two variables of described in FIG. 4 and FIG.
a predicted bit length operation unit 4 containing a register A and a register B for holding s and width, an input / output control signal generation unit 5 for controlling input / output, and a memory for holding a constantly changing width value Stack memory 6
And a run of a predicted bit length and a predicted bit length / dominant symbol setting unit 7 for setting a superior symbol.

The buffer register 2 has a capacity to store at least the input bit sequence 1 having the number of bits corresponding to the maximum predicted bit length run set by the predicted bit length / dominant symbol setting unit 7. Also, it has a function of inverting and inputting the input bit sequence 1 according to an instruction of the predicted bit length / dominant symbol setting unit 7.

The determination section 3 determines the position ofs on the buffer register 2 output from the prediction bit length calculation section 4 and the ofs of the position ofs.
, The input bit sequence 1 on the buffer register 2 determined by the width indicating the number of bits to be predicted is selected, and when they are all “0”, “0” is set as the code bit 10, that is, “0” And outputs "1" as a misprediction signal when "1" is included.

The prediction bit length calculation unit 4 performs the
A register A and a register B that hold two variables ofs and width are incorporated. Of these variables, width
Is set to the bit length run for primary prediction output from the prediction bit length / dominant symbol setting unit 7 in response to the encoding start signal input from the input / output control signal generation unit 5, while ofs is 0 Is initialized to

After this initialization, from the next cycle, the prediction bit length calculation unit 4 causes the determination unit 3 to check the logic of the input bit sequence 1 on the buffer register 2 determined by the ofs and the width, and to determine the determination result. take in. When the determination result is "1", that is, when the prediction is incorrect, the width of the register B is reduced to 1/2 and the value is written to the stack memory 6. However, this does not apply when the width is already 1, and in this case, the width of the register B, that is, 1 is added to the ofs of the register A, and the value of the width written immediately before from the stack memory 6 is read. , Stored in the register B.

When the judgment result is "0", that is, when the prediction is correct, the width of the register B is added to the ofs of the register A, and the value of the width written immediately before from the stack memory 6 is added. Read and store in register B. The above-described processing is performed while determining the input bit sequence 1 for each cycle. Further, as the above-described encoding process proceeds, the ofs of the register A gradually increases. When this ofs becomes equal to the temporary prediction run, the encoding of the input bit sequence 1 fetched into the buffer register 2 is performed. Complete. Therefore, at this time
An input / output control signal generator 5 sends a signal indicating completion of encoding.
Then, the input / output control signal generator 5 receives this, activates the input bit request signal 8, and simultaneously instructs the buffer register 2 to take in a new input bit sequence 1.

The input / output control signal generator 5 receives the input bit request signal 8 At the same time, and instructs the buffer register 2 to take in the number of input bit sequences 1 indicated by the prediction run inputted by the prediction bit length and dominant symbol setting unit 7. Then, when the number of bits indicated by the prediction run is input to the buffer register 2, the encoding start signal is output to the prediction bit length calculator 4. Also, the strobe signal 9 serving as a synchronization signal is activated until the encoding is completed with a delay of one cycle from the output thereof.

The stack memory 6 is a memory for holding the value of the width that changes every moment as described above, and is a so-called first-in last-out memory. That is, it is a memory that outputs the value written first at the end, or conversely, a memory that outputs the value written most recently at the beginning. The value of the width that changes every moment is written in the stack memory 6.

The predicted bit length / dominant symbol setting unit 7
From the prediction bit length calculation unit 4, the past coding state (how much the prediction was correct or how much the prediction is incorrect, etc.) is input, and the prediction bit length run of the newly input bit sequence 1 and the superiority Set the symbol. Prediction bit length r
un is input to the prediction bit length calculator 4, and the dominant symbol setting signal is input to the buffer register 2.

The binary bit string encoding process described above is executed by the binary bit string data encoding apparatus having the above configuration. FIG. 7 shows the compression ratio and the encoding time of this encoding process. FIG. 7 shows the compression ratio and the encoding time when the binary bit string encoding process used in the present invention is used for the four types of files, and also shows the compression ratio and the encoding of the conventional QM coder for reference. The time is also shown. As shown in FIG. 7, the encoding method of the binary bit string used in the present invention is as follows.
The compression ratio is at the same level as the QM coder, and the encoding time is greatly reduced.

Next, the process of decoding a binary bit string according to the present invention will be described. The decoding process of the binary bit string includes a decoding main routine of FIG. 8 and a decoding subroutine of FIG. The decoding subroutine of FIG. 9 is recursively called in the same routine as the encoding subroutine of FIG.

First, each step of the decoding main routine of FIG. 8 will be described. First, setting of an initial value run of prediction and selection of a dominant symbol are performed (step S30).
These values are naturally set to the same values as those on the encoding side.
Next, 0 is set to the local variable ofs, and run is set to the width.
Is substituted (step S31). Then, one bit of a code word to be the code bit output 10 is input (step S3).
2). Next, in step S33, the logic of the code word is checked. If "0", the process proceeds to step S34, and if "1", the process proceeds to step S35. Here, the code word “0” is a signal per prediction indicating that the prediction was successful.

If the codeword is "0", the dominant symbol is written from A [ofs] to A [width-1] (step S34). On the other hand, the code word “1” indicates that the prediction has been lost. Then, at this time, first, wid
Check the size of th and divide it if it exceeds 1,
Since re-prediction is possible, the flow advances to step S37 to call a decoding subroutine. If it is 1 or less, proceed to step S36,
Write the inferior symbol to A [ofs].

In step S38, the predicted run is reset and, if necessary, the dominant symbol is changed. The resetting and changing are naturally performed according to the same rules as those on the encoding side. Subsequent to step S38, in step S39, decoded data of A [ofs] to A [width-1] is output.

If the primary prediction at step S33 in the decoding main routine shown in FIG. 8 is incorrect, the decoding subroutine shown in FIG. 9 is called at step S37. The argument passed to this decryption subroutine is o
fs and width. Hereinafter, each step of the decoding subroutine will be described.

First, the prediction range is divided into two by setting the width to １ ／ (step S40). Next, one bit of a code word is input (step S41). This bit is the prediction result of the first half sequence. In the next step S42, the logic of the input code word is checked. If "0", the process proceeds to step S43, and if "1", the process proceeds to step S45. Here, the codeword “0” indicates that the prediction was successful. Therefore, the dominant symbol is written into the first half sequence, that is, from A [ofs] to A [ofs + width−1] (step S43). Next, in step S44, the pointer ofs indicating the head position of the first half series is set to wi.
dth is added to change to indicate the head position of the latter half series. In addition, when the prediction of the first half sequence is correct, the inferior symbol always exists in the second half sequence. Therefore, step S49 for making a prediction determination of the second half series and step S49
The process skips 50 and proceeds to step S52.

On the other hand, when the code word is "1", the step S
The process proceeds from S42 to step S45, and the width is checked. When the value is 1 or less, it is not necessary to perform the division prediction any more, that is, since the decoded bit is determined, the process proceeds to step S46. In the case of two or more, since it is necessary to further divide and predict, a decoding subroutine is called in step S47. In step S46, the inferior symbol is written to A [ofs]. In step S47, the decoding subroutine is recursively called as a child process.

[0097] Subsequent to step S46 or step S47, step S48 is executed. This step S4
In 8, the width is added to the pointer ofs indicating the head position of the first half series, and the pointer is changed to indicate the head position of the second half series. Next, a 1-bit code word is input (step S49). This bit is the prediction result of the latter half sequence.
In step S50, the logic of the input code word is checked. If “0”, the process proceeds to step S51, and if “1”, the process proceeds to step S52.

Here, the code word "0" indicates that the prediction was successful. Therefore, the second half sequence, that is, A [of
s] to A [ofs + width−1] to input a dominant symbol (step S51). On the other hand, if the code word is "1"
In step S52, the flow advances to step S52 to check the width. If it is equal to or less than "1", it is not necessary to divide and predict any more, that is, the decoded bit is determined. In the case of two or more, since it is necessary to further divide and predict, a decoding subroutine is called as a child process in step S54.

Next, the decoding process of the binary bit string will be described based on a specific example. That is, 2
“10110” obtained in a specific example of encoding of a coded bit string
The decoding process will be specifically described using the code bit output 10 of 10 ″. As in the case of the encoding, the decoding is executed with the initial value of prediction run being 8 and the dominant symbol being 0.

First, in step S32, the first code bit is input. Since this is "1", the flow advances to step S35 to check the width. At this time wi
Since dth is 8, the decryption subroutine is called in step S37. In this decoding subroutine, after changing the width to 4 in half in step S40, the second code bit is input in step S41. This is "0"
Therefore, the process proceeds to step S43, where the first half series, that is, A

"0" is written from [0] to A [3]. After changing ofs to 4 in the following step S44, step S5
The process proceeds from Step 2 to Step S54. In this step S54, a child decoding subroutine is called as a child process.

In the child decoding subroutine, step S
After the width is further halved at 40 and changed to 2, the third code bit is input at step S41. Since this is "1", the process proceeds to step S47 via step S45.
Then, a grandchild decoding subroutine is called as a grandchild process. In this grandchild decryption subroutine, step S40
After the width is changed to 1 in step S4, 4 is set in step S41.
Enter the th sign bit. Since this is "1", the process proceeds to step S45. At this time, since the width is 1, the process proceeds to step S46, where 1 is written to A [4].
Subsequently, after changing ofs to 5 in step S48, the fifth code bit is input in step S49. Since this is "0", the process proceeds to step S51, where "0" is written to A [5].

Thereafter, the process returns to the child decoding subroutine, and ofs is changed from 4 to 6 in step S48. In a succeeding step S49, a sixth sign bit is inputted.
Since this is "1", the process proceeds to step S54 via step S52, where the grandchild decoding subroutine is called again as a grandchild process. The grandchild decryption subroutine
First, after the width is changed to 1 in step S40,
In step S41, the seventh code bit is input. Since this bit is "0", 0 is written to A [6] in step S43. In the following step S44, ofs is changed to 7,
In step S52, since the width is "1", the process proceeds to step S53, where 1 is written to A [7].

Thereafter, the process returns to the child decoding subroutine and the decoding subroutine, and further returns to the decoding main routine. In the restored decoding main routine, the prediction run is reset and the dominant symbol is reset in step S38. Then, in step S39, the decrypted data is

Output from [0] to A [7].

The decoded data obtained by the above-described decoding process of the binary bit string is “000010001”, and the original input bit sequence 1 is restored.

Next, a binary bit string data decoding apparatus for performing the above-described binary bit string decoding process will be described with reference to FIG.

The binary bit string data decoding apparatus includes a predicted bit length run and a predicted bit length / dominant symbol setting unit 11 for setting a superior symbol, a register A for holding two variables ofs and width, and A prediction bit length calculation unit 12 having a register B therein, a stack memory 13 as a memory for holding an ever-changing width value, an input / output control signal generation unit 14 for controlling input / output, and decoding It mainly comprises a decoding bit setting unit 15 for setting bits.

Here, the prediction bit length / dominant symbol setting unit 11 receives from the prediction bit length calculation unit 12 the past decoding state (how much prediction has been achieved or how much the prediction has deviated, etc.). And newly input sign bit 1
A predicted bit length run of 0 and a dominant symbol are set. The predicted bit length run is input to the predicted bit length calculation unit 12, and the dominant symbol setting signal is input to the decoded bit setting unit 15.

As described above, the prediction bit length calculation unit 12 has a built-in register A and a register B for holding two variables ofs and width. widt
In h, the bit length run for primary prediction set by the prediction bit length / dominant symbol setting unit 11 is set, and ofs
Is initialized to 0. Register B, ie width
Is supplied to the input / output control signal generator 14. On the other hand, after the initialization, when the input code bit 10 is “1” and the width is 2 or more, the width is halved,
Write to the stack memory 13. However, width
Is 1, a decoded bit output permission signal is output to the decoded bit setting unit 15 and the input / output control signal generating unit 14. At the same time, the width of the register B is added to the ofs of the register A, and Wid written
th is read and held in the register B.

When the sign bit 10 is "0", the decoding bit output enable signal is output to the decoding bit setting unit 15 and the input / output control signal generation unit 14, and at the same time,
After adding the width of the register B to s, the width written immediately before is read from the stack memory 13 and held in the register B. Then, ofs gradually increases as decoding progresses. Therefore, when ofs and run become equal, a signal indicating that decoding of the code bit 10 of one prediction unit is completed is generated, and this signal is output to the prediction bit length / dominant symbol setting unit 11. Receiving this signal, the predicted bit length / dominant symbol setting unit 11 resets the predicted bit length run for the next code bit 10 and resets the superior symbol.

The stack memory 13 is a memory for holding the value of the width that changes every moment as described above, and is a so-called first-in last-out memory. That is, it is a memory that outputs the value written first at the end, or conversely, a memory that outputs the value written most recently at the beginning.

The input / output control signal generator 14 outputs the strobe signal 1 indicating that the valid decoded bit 20 is being output.
6 is generated. Predicted bit length calculator 12
Receiving the decoded bit output permission signal,
The strobe signal 16 is output for a period specified by dth. When the strobe signal 16 is active for two or more cycles, the code bit request signal 17 is made non-active from the second cycle to suppress the input of a new code bit 10 and, at the same time, the predicted bit length calculation unit 1
Operation 2 is also temporarily stopped.

The decoding bit setting section 15 is constituted by a register which receives the decoding bit output permission signal from the prediction bit length calculating section 12 and temporarily holds the code bit 10 being input at that time. In addition, the decoding bit 20 is inverted as required according to the instruction of the prediction bit length / dominant symbol setting unit 11.

The above-described decoding process is executed by the binary bit string data decoding apparatus having the above-described configuration. The decoding time when this decoding process is used is much shorter than that of the QM coder, like the encoding time shown in FIG. That is, since the decoding process of the binary bit string uses an algorithm reverse to that of the encoding process, if the encoding time is reduced, the decoding time is also reduced.

The data encoding and decoding of such a binary bit string can be said to be, so to speak, a predictive run-length encoding method and a decoding method. Then, the predictive run-length coding (Predictive Run
-Length Coding and translated into English
C) can be applied to a multi-valued information source, and its application will be described below.

In general, the appearance probabilities of “0” and “1” are different for each bit plane in a multi-valued information source. Therefore, efficient encoding cannot be performed.
Thus, multi-value prediction run-length coding (hereinafter multi-value PRL)
C) is divided into bit planes, that is, in the case of, for example, an 8-bit multi-valued information source, divided into eight bit planes, and the data of each bit plane is divided. It is conceivable that each is coded, and the simulation shows high coding efficiency.

This multi-valued P which is divided into bit planes
In the first method of RLC, although high coding efficiency is indicated, correlation between planes is ignored.
The efficiency is slightly lower than when using a QM coder. Therefore, when considering a case where each value of the multi-valued information source is concentrated on lower bits, the following second method of the multi-valued PRLC can be considered.

The second method of the multi-value PRLC is to apply the PRLC for each bit plane similarly to the first method.
C is performed. Then, when "1" appears, the subsequent lower bits are directly output to the stream. This second method will be described based on the flow diagram of FIG. 11 and the specific example of FIG.

For example, as shown in FIG. 12, input symbols serving as multi-value information sources are "0", "1", "4", "9", and "2".
In the case where “3”, “60”, “224”, “0”, and “3” continue, the most significant bit plane is “00000000100...”. Then, the uppermost bit plane is coded by the above-described PRLC with a predetermined prediction bit length run (step S55). At this time, it is determined whether or not "1" appears (step S56). For the "224" input symbol in which "1" appears, the value of the lower seven bit planes "1100000" is directly input. The data is output to a stream (step S57) and encoded.

Next, for the bit whose most significant bit is not "1", the second bit is subjected to PRLC (step S5).
8). In the specific example of FIG. 12, "00000000
... "is to be PRLC. At this time, "224"
The second bit "1" is skipped because it has already been output. Further, the number of predicted bit lengths run when performing PRLC is a value determined independently and independently of the number of predicted bit lengths used for the most significant bit. Here, it is detected whether or not the second bit has "1" (step S59). If "1" is present, the lower 6 bits are output to the stream (step S60).
In the specific example of FIG. 12, since there is no “1”, step S6
0 is not performed, and the next third bit is subjected to PRLC with its own prediction bit length run (step S61).

Then, it is confirmed whether or not "1" is present in the third bit (step S62). At this time, the third bit is “00000100...” And the symbol “6
For "0", the lower 5 bits "11100"
Is output to the stream (step S63). afterwards,
Although the third bit is not "1", the fourth bit is subjected to PRLC with a unique prediction bit length run (step S6).
4). Then, it is determined whether the fourth bit has "1" (step S65). At this time, the fourth bit is “00”.
00100..., And for the symbol “23”, the lower 4 bits “0111” are output to the stream as it is (step S66).

The fifth bit,
The sixth bit, the seventh bit, and the least significant bit are subjected to PRLC with their own predicted bit lengths run (steps S67 to S76). In the example of FIG. 12, in the least significant bit PRLC of step S76, “0” of the input symbol “0” and “1” of the subsequent “1” are targets, and the subsequent input symbols “4”, “9”, and “9” 23 ”,“ 6 ”
The least significant bits of “0” and “224” have already been output and are out of scope. Then, “0” of the following “0” is targeted, and the next “3” is not targeted. Therefore, step S
At 76, "010 ..." is the target of PRLC. Note that when encoding each bit plane, the number of bits in the buffer that satisfies the corresponding condition depends on the predicted bit length run.
If the number of symbols is less than the predetermined value, encoding is performed by inserting a dummy dominant symbol in an insufficient amount. In this case, on the decoding side, an extra dominant symbol is decoded, but if the capacity of the buffer at the time of encoding is known, whether the decoded bits are significant or not.
It is possible to determine whether a dummy has been inserted. In order to avoid such a method of using the dummy dominant symbols, it is sufficient to make the buffer large enough or to have a capacity enough to store the entire file to be compressed.

Next, a description will be given of a third method of multi-valued PRLC in which further improvement in efficiency can be expected. This uses the fact that PRLC and the QM coder have little performance as the compression ratio when the information source is treated as non-memory as described above. That is, in PRLC,
The focus is on the inference that if the Markov model used in the QM coder is well adapted, a compression rate similar to that of the QM coder should be obtained. In the case of the QM coder, the probability estimation table is subtracted according to the context, and the occurrence probability of the inferior symbol is obtained. If this is applied to PRLC, a plurality of sequences are created by classifying the coded symbols for each context, and This is equivalent to separately performing PRLC. In other words, if the coding is performed separately on the level plane (= one corresponding to each different value of the multi-valued information source) instead of the bit plane, the compression efficiency is further improved.

That is, the appearance rate of each input symbol is biased, and this bias increases the compression efficiency when encoding each input symbol, that is, the level plane. For example, in the case of a 4-bit input symbol, there are a total of 16 types of input symbols, of which "0001" is 1 with a probability that "0001" is close to 0.
It is assumed to occur with a probability close to. Then, "000
The input symbol “1” has many occurrences of “0”, while the input symbol “1010” has many occurrences of “1”. As a result, PRLC
The compression ratio is increased by using.

The third method of the multi-valued PRLC described above does not matter when the number of bits of the input symbol is small, but when the number is large, the memory becomes large and poses a problem. For example, when the input symbol is 8 bits, the number of level planes is 25.
6, the management of the prediction bit length run for each plane and the prediction management of how many input symbols are stored in the buffer become difficult, the memory for the management becomes large, and the device becomes large. Increase costs. Next, a fourth method of the multi-value PRLC that solves this problem will be described.

First, it is assumed that the input symbols to be input are converted by the head moving dictionary method so that the appearance probabilities are biased. That is, it is biased toward lower bits or upper bits. Other factors that bias the appearance probability include quantization coefficients such as DPCM and IHT. In any case, the input symbols are concentrated on the lower bits, the upper bits, or the like, or the input symbols concentrated in such a manner are handled. Then, only the level planes with a high concentration of probabilities are handled independently, and those with a low concentration of probabilities are handled collectively with a plurality of level planes. This prevents the memory from becoming large, while minimizing the reduction in compression efficiency.

A fourth method of the multi-value PRLC will be described with reference to FIGS. This specific example is suitable for a case where input symbols are concentrated on lower bits, as in the second method. In the fourth method of multi-value PRLC, input symbols are first divided into groups. For example, in the case of an 8-bit input symbol, it is divided into nine groups as shown in FIG. Then, this group number (specifically, a bit for determining whether or not the group corresponds to a group described later) is subjected to PRLC. A specific method of this PRLC will be described based on a flow diagram of FIG. 14 and a specific example of FIG.

For example, as shown in FIG. 15, input symbols serving as multi-value information sources are "0", "1", "4", "9", and "2".
Considering the case where “3”, “60”, “224”, “0”, and “3” continue, the first input symbol “0” falls into the group number “0”. The determination bit of whether or not it is set to “0” to the effect. Also, since “1” is not the group number “0”, the determination bit of the group number “0” part is “1”, and the determination bit of the group number “1” is “0”. The subsequent determination bit of “4” is “1110”, the determination bit of “9” is “11110”, the determination bit of “23” is “111110”, the determination bit of “60” is “1111110”, and “ The determination bit of “224” is “111111110”.

The above decision bits are subjected to PRLC. In this specific order, first, a determination bit indicating whether the input symbol corresponds to the group number “0” is PRLC (step S77). In the example of FIG. 15, the determination bit string “011111101”
... "Are PRLCed with a unique prediction bit length run. Then, it is determined whether or not there is a group corresponding to the group number “0” (step S78). If there is “0”, the value is the input symbol “0”, and the encoding is completed.
Thus, the encoding of the first and eighth input symbols “0” is completed. If there is a group number other than "0", PRLC is performed on a determination bit as to whether or not the group number is "1" (step S79). FIG.
In the example of No. 5, the second determination bit string “011”
., 1111... With a unique prediction bit length run.

Then, it is determined whether or not there is a group whose group number corresponds to "1" (step S80). If there is one corresponding to the group number “1”, the encoding is completed. Here, since the corresponding code is “0”, the second input symbol “1” corresponds. Therefore, the encoding of the input symbol “1” is completed.

Next, if there still exists a group whose group number is not “1”, the judgment bit indicating whether there is a group corresponding to the group number “2” is PRLC (step S81). In the example of FIG. 15, the input symbol "4"
The sequence of decision bits “111110...” Corresponding to “9”, “23”, “60”, “224”, and “3” is subjected to PRLC with a unique prediction bit length run. Then, it is confirmed whether or not there is a group whose group number corresponds to "2" (step S82). If there is one corresponding to the group number “2”, one additional bit “0” or “1” is output to the stream (step S83), and the encoding is completed. In the example of FIG. 15, the ninth input symbol “3” corresponds, and the additional bit “1” is output to the stream. When the input symbol is “2”, the group number is the same as the input symbol “3”, which is “2”, but the additional bit is “0”.

Next, if there is still a group whose group number is not "2", a judgment bit indicating whether there is a group corresponding to the group number "3" is PRLC (step S84). In the example of FIG. 15, the determination bit string “011
11... With a unique prediction bit length run.
Then, it is determined whether or not there is a group corresponding to the group number “3” (step S85). If there is one corresponding to the group number "3", one of the two additional bits "00", "01", "10", and "11" is output to the stream (step S86). In the example of FIG.
"3" corresponds to the third input symbol, and an additional bit "00" is output to the stream. The additional bit is "01" when the input symbol is "5", "10" when it is "6", and "11" when it is "7".

The processing is executed according to a similar flow (steps S87 to S99). That is, the determination bits indicating whether or not there are groups corresponding to the group numbers “4”, “5”, “6”, and “7” are respectively set in PRLC.
(Steps S87, S90, S93, S96) and if there is a corresponding number, additional bits of 3, 4, 5, and 6 bits are output to the stream (steps S89, S92, S95, S98). .
Finally, if there is a group other than the group number "7", 7 additional bits are output to the stream (step S99), and one cycle of encoding is completed.

The processing speed of the fourth method of the multi-value PRLC is higher than that of the second method. That is, in the fourth method, when the group number is determined, the encoding of the higher-order determination bits is not performed. On the other hand, in the fourth method, as in the second method, when encoding the determination bits of each group number, the number of bits in the buffer that satisfies the corresponding condition is equal to the predicted bit length ru.
If n is less than n, encoding is performed by inserting a dummy dominant symbol by the shortage. In this case, on the decoding side, an extra dominant symbol is decoded, but if the capacity of the buffer at the time of encoding is known, it is possible to determine whether the decoded bit is significant or inserted as a dummy. . In addition,
In order to avoid such a method of using the dummy dominant symbol, the buffer may be made sufficiently large, or may have a capacity sufficient to store the entire file to be compressed.

Data obtained by comparing each of the above methods with the conventional method are shown in Tables 1 and 2 below and FIGS. 16 and 17. FIG. 16 is a graph of Table 1 showing a comparison of compression ratios, and FIG. 17 is a graph of Table 2 showing a comparison of decoding processing speeds.

[0135]

[Table 1]

[0136]

[Table 2]

The source file for comparison is 25 files.
A multi-valued information source is used in which a bitmap file of a six-color image is index-converted by a head moving dictionary. Here, the leading moving dictionary is a 256-entry dictionary of 256 entries.
In this method, the dictionary is switched according to the immediately preceding coded symbol, and the leading movement of the dictionary is 先頭 leading movement.
The method adopted as the conventional method is that the 8-bit index is directly encoded by the QM coder.
M "," AC "in which multi-valued data is encoded as it is by arithmetic coding, and" Huf using two-pass Huffman.
fman ", which is adopted as the method of the present invention is" TM8 "of the second method of the multi-valued PRLC described above.
E4 "and" TM8E5 "of the fourth method.

[0138] In both "TM8E4" and "TM8E5", the initial value of the predicted bit length run is 1 and the value of the maximum predicted bit length run is 256. Then, the condition for doubling the prediction bit length run is when two or more predictions are hit consecutively, and the condition for halving the run is when the prediction is missed. Further, the condition for changing the dominant symbol is that the prediction bit length run is 1 and the prediction is incorrect. As for the compression ratio, as shown in Table 1 and FIG. 16, "TM8E4" and "TM8E5" achieve the same compression efficiency as the QM coder. In particular, "TM8E
“5” indicates that up to three of the four evaluation files exceed the compression rate of the QM coder. The processing time required for decoding is “TM8” as shown in Table 2 and FIG.
E4 "is about 1/2 of the QM coder, and" TM8
E5 "is about 1/3 of the QM coder. In particular, the file BE256. In BMP,
“TM8E5” achieves a processing speed comparable to “Huffman”. Note that “TM8E4” and “TM8E
In a PLRC such as "5", when the prediction bit length is relatively small, such as 2, 4, and 8, coding and decoding can be performed by creating a code table in advance, as described later. An improvement in processing speed can be expected. Furthermore, if the control of the prediction bit length run is optimized, the efficiency can be further increased.

Each of the above embodiments is an example of a preferred embodiment of the present invention. However, the present invention is not limited to this, and various modifications can be made without departing from the scope of the present invention. It is. For example, the codeword to be output when the prediction is successful is set to "1" instead of "0", and when the prediction is incorrect, it is set to "0" instead of "1". It is also possible to output the inferior symbol when the prediction is incorrect. Further, the new reduced predicted bit number can be set to 1/3, 1/4 or the like instead of 1/2 of the original predicted bit number, or a number obtained by subtracting a predetermined number from the original predicted bit number. On the other hand, the new increase prediction bit number is not twice the original prediction bit number, but may be tripled or quadrupled, or may be a number obtained by adding a predetermined number to the original prediction bit number. Note that the new increase prediction bit number is not limited, and a predetermined value, for example, 256 bits or a multiple of 2 may be set as the maximum value. Also, the minimum value of the new decrease prediction bit number is not 1, but
Other numerical values such as 2 and 3 may be used.

Also, in the division encoding step, the division is
Instead of the division, the target sequence may be divided into three, four, or the like, or the first half target sequence and the second half target sequence may be unequally divided. Furthermore, instead of performing recursive repetition such as division when prediction is missed until the number of bits becomes 1, the number of bits becomes 2
The processing may be terminated when the number reaches 3 or 3, and thereafter, the input bit sequence 1 may be output as it is, or may be encoded by a QM coder.

Further, in the above-described embodiment, when the prediction is incorrect, first, the first half is coded by paying attention to the first half, and then the second half is coded. May be used to perform encoding and decoding. That is, encoding and decoding may be performed with priority given to the latter half. Further, the encoding and decoding may be performed by setting the priority portion to the first half or the second half, or by appropriately mixing them.

Further, in the above-described embodiment, all the noticed sequences determined by the number of prediction bits run are predicted as dominant symbols, and if the prediction is successful, "0" or the like is output as a codeword, Outputs "1" and the like, further divides the sequence into two parts, performs the same operation until the prediction is correct or the division becomes impossible, and performs encoding. However, if the number of predicted bits is relatively short, for example,
In the case of 1, 2, 4, 8, etc., an encoding table is prepared in advance, and by referring to the table, 1
Encoding may be performed in cycles. For example, when the number of predicted bits is 2, the dominant symbol is “0”, when the input bit sequence 1 is “00”, the sign bit 10 is “0”, and when it is “01”, it is “10”. Since "10" is "110" and "11" is "111", this relationship is prepared in advance as an encoding table. By using such a method, the encoding time can be further reduced. In this case, if an encoding table is selected and encoded according to the number of prediction bits run, code bits 10 similar to those in the above-described embodiment can be obtained. Also,
Similarly, in decoding, by preparing a decoding table and referring to the table selected by the predicted bit number run, decoding can be performed with the decoding time reduced.

As described above, the application of the present invention to a multi-valued information source is not limited to the case of data encoding, and it goes without saying that
The same algorithm can be applied to data decoding. Also, the second and fourth methods of multi-value PRLC can be applied to ordinary run-length coding instead of PRLC.

[0144]

As described above, in the data encoding method and data encoding apparatus for a multilevel information source according to the present invention, Q
While coding efficiency comparable to that of the M coder is obtained, the coding speed is much higher than that of the QM coder. For this reason,
It is the most practical in terms of the compression method of various multi-valued information sources currently used.

Further, in the data decoding method and the data decoding apparatus for a multi-valued information source according to the present invention, a decompression efficiency similar to that of a QM coder can be obtained, but the decoding speed is much higher than that of a QM coder. Becomes For this reason, it is the most practically practical decoding method among various multi-valued information sources currently used, and the convenience is improved.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining an outline of an algorithm for predictive run-length encoding used in the present invention, and is a diagram illustrating a relationship between a sequence of interest and a predicted bit number run.

FIG. 2 is a diagram for explaining an outline of an algorithm for predictive run-length coding used in the present invention, and is a diagram showing a state in which a sequence of interest in FIG. 1 is divided.

FIG. 3 is a diagram for explaining an outline of a prediction run-length encoding algorithm used in the present invention, and is a diagram showing a state in which the first half attention sequence in FIG. 2 is further divided.

FIG. 4 is a flowchart for explaining an encoding process of predictive run-length encoding used in the present invention, and is a flowchart showing an encoding main routine.

FIG. 5 is a flowchart for explaining an encoding process of predictive run-length encoding used in the present invention, and is a flowchart showing an encoding subroutine.

FIG. 6 is a block diagram showing a configuration of an embodiment of a data encoding device for predictive run-length encoding used in the present invention.

FIG. 7 is a diagram illustrating a data encoding method of prediction run-length encoding used in the present invention and a compression rate and an encoding time by the data encoding device.

FIG. 8 is a flowchart for explaining a decoding process of predictive run-length encoding used in the present invention, and is a flowchart showing a decoding main routine.

FIG. 9 is a flowchart illustrating a decoding process of predictive run-length encoding used in the present invention;
It is a flowchart which shows a decoding subroutine.

FIG. 10 is a block diagram showing a configuration of an embodiment of a data decoding device for predictive run-length encoding used in the present invention.

FIG. 11 is a flow diagram showing a second method of encoding data of a multi-level information source according to the present invention.

FIG. 12 is a diagram for explaining a specific operation of the second method shown in FIG. 11;

FIG. 13 is a diagram for explaining grouping in a fourth method of the data encoding method of the multilevel information source according to the present invention.

FIG. 14 is a flow diagram illustrating a fourth method of the multi-level information source data encoding method according to the present invention.

FIG. 15 is a diagram for explaining a specific operation of the fourth method shown in FIG. 14;

FIG. 16 is a diagram showing a compression ratio of a data encoding method using a second method and a fourth method of a multi-valued information source according to the present invention, and comparing with a conventional method.

FIG. 17 is a diagram showing processing time required for decoding of the data encoding method of the multi-valued information source using the second method and the fourth method according to the present invention, and comparing with the conventional method. .

FIG. 18 is a diagram illustrating a configuration of a QM coder which is a conventional arithmetic code type entropy encoder.

FIG. 19 is a flowchart showing the operation of the QM coder in FIG. 18;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input bit sequence 2 Buffer register 3 Judgment part 4 Prediction bit length calculation part 5 I / O control signal generation part 6 Stack memory 7 Prediction bit length and dominant symbol setting part 8 Input bit request signal 9 Strobe signal 10 Sign bit 11 Prediction bit Dominant symbol setting unit 12 Predicted bit length calculation unit 13 Stack memory 14 I / O control signal generation unit 15 Decoding bit setting unit 16 Strobe signal 17 Code bit request signal 20 Decoded bit run Predicted bit number

Claims (24)

[Claims] 1. A multi-valued information source composed of a plurality of bits is divided into bit planes, and when a binary bit string composed of “0” and “1” of each bit plane is inputted, “0” is input.
A prediction setting step of predicting that one of “1” is a dominant symbol and the other is a inferior symbol, predicts that the dominant symbol is continuous n, and sets the n as the number of prediction bits; When prediction is performed on the target sequence composed of the predicted number of bits, either one of "0" and "1" is output as a codeword as a signal per prediction, and the next n bit strings are coded. A prediction result output step of outputting the other signal of either “0” or “1” as a code word as a prediction error signal when the prediction is deviated, and the prediction bit number when the prediction is deviated a predetermined number of times. Wherein the same prediction setting step and prediction output step are recursively repeated with n being the new reduced prediction bit number less than n. 2. The method according to claim 1, wherein said n number is an even number, and said sequence of interest is divided into two when said prediction is deviated by a predetermined number of times, and when there is a inferior symbol only in the first half of the sequence of interest. , The new reduced prediction bit number is set to の of the prediction bit number, and when the inferior symbol exists in the latter half target sequence of the second half, the new reduction prediction bit number is set to 1 / １ ／ of the prediction bit number. 4. The data encoding method for a multi-valued information source according to claim 1, wherein the number is set to 4. 3. When the new reduced prediction bit number is 1 and the bit is a inferior symbol, in the subsequent encoding, the conventional inferior symbol is set as the superior symbol and the conventional superior symbol is encoded as the inferior symbol. 3. A data encoding method for a multi-valued information source according to claim 1, wherein: 4. The data encoding method according to claim 1, wherein the predetermined number of times is one. 5. The multi-valued information according to claim 1, wherein the number of predicted bits is set to a new increased predicted bit number larger than n when the predetermined number of predictions are performed. Source data encoding method. 6. A multi-valued information source composed of a plurality of bits is divided into bit planes, and when a binary bit string consisting of “0” and “1” of each bit plane is input, “0” is input.
A prediction setting step of predicting that one of “1” is a dominant symbol and the other is a inferior symbol, predicts that the dominant symbol is continuous n, and sets the n as the number of prediction bits; When prediction is performed on the noticed sequence consisting of n bit strings, one of "0" and "1" is output as a codeword as a signal per prediction, and the next n bit strings are encoded. Work, and when deviated, codeword "0"
Or a prediction result output step of outputting either one of the signals “1” as a non-predicted signal, and a similar prediction in which the number of predicted bits is set to a new increased predicted bit number larger than n when the prediction is performed a specified number of times. A data encoding method for a multi-valued information source, wherein a setting step and a prediction result output step are repeatedly performed. 7. The data encoding method for a multi-valued information source according to claim 5, wherein the prescribed number of times is set to two times, and the new increase prediction bit number is set to twice the prediction bit number. . 8. The data encoding method according to claim 1, wherein n is 2 ^m (m is an integer of 0 or more). 9. When the prediction is deviated, the sequence of interest is divided into two, and when the first half sequence of interest in the first half is entirely the dominant symbol, “0” is output as a codeword, The second half of the generated second half sequence of interest is further divided into two, and a codeword of “0” or “1” is output, and a codeword is generated when the inferior symbol exists in the first half of the sequence of interest. And outputs “1” as
The first-half target sequence is further divided into two, and a first-half departure step of outputting a code word of “0” or “1” is provided.
9. The method according to claim 8, wherein, as long as the inferior symbol exists in each divided sequence of interest, division of the sequence of interest is repeated, and the first half hitting step and the first half separating step are recursively repeated. Data encoding method for value information source. 10. When the prediction is deviated, the target sequence is divided into two, and when all the latter half target sequences of the latter half are the dominant symbols, “0” is output as a codeword, and The first half attention series of the first half is further increased by 2
The latter half step of dividing and outputting a code word of “0” or “1”; and outputting the code word “1” when the inferior symbol exists in the latter half notice sequence, The target sequence is further divided into two, and the latter half-off step of outputting a code word of “0” or “1” is performed. As long as the inferior symbol exists in each divided target sequence, the division of the target sequence is performed. 9. The data encoding method for a multi-level information source according to claim 8, wherein the step of repeating the latter half part and the step of separating the latter half part are recursively repeated. 11. A multi-valued information source consisting of a plurality of bits is divided into bit planes, and the bit plane of the most significant bit is encoded by the data encoding method according to any one of claims 1 to 10, wherein "1" A data encoding method for a multi-valued information source, wherein the following lower bits are output as code bits when "" appears. 12. A multi-level information source which divides a multi-level information source consisting of a plurality of bits into bit planes and compresses and encodes a binary input bit sequence consisting of “0” and “1” of each bit plane In the data encoding device of the information source,
Either “0” or “1” is set as a dominant symbol, and the other is set as a dominant symbol, and it is predicted that n dominant symbols are continuous, and the set n is set as the number of bits to be predicted. An operation setting unit, a buffer register for temporarily storing an input bit sequence, and input of each value of the operation setting unit and the buffer register for the prediction bit length, etc., and a prediction is performed for a target sequence consisting of the input prediction bit number. Either "0" or "1" is output as a per-prediction signal as a codeword when hit, and "0" is output as a codeword when deviated.
Or a determination unit that outputs the other signal of “1” as an out-of-prediction signal, and when the prediction deviates by a predetermined number of times, sets the predicted bit number smaller than n to the new reduced predicted bit number Set in the length calculation setting section,
A data encoding device for a multi-valued information source, wherein when the number of predictions reaches a prescribed number of times, the number of predicted bits is set to a new increased predicted bit number greater than n by the predicted bit length etc. operation setting unit. . 13. A multi-valued information source comprising a plurality of bits is divided into bit planes, and a buffer register comprising a group of registers for temporarily storing an input bit sequence of each bit plane, and a head position of a sequence of interest to be encoded are defined. The variable of which to indicate
register holding s and variable w indicating the predicted bit length
a prediction bit length calculation unit including a register for holding the id, and an input bit sequence on the buffer register determined by the width based on the position of ofs on the buffer register output from the prediction bit length calculation unit and the ofs thereof A determination unit that outputs one superior symbol as a code bit output when all the input bit sequences are superior symbols, and outputs one inferior symbol when the inferior symbol is included, and an input bit sequence in the buffer register. And an input / output control signal generator for requesting an input bit when receiving a completion signal from the prediction bit length calculator from the prediction bit length calculator,
stack as a memory for holding the value of th
A memory, and a prediction bit length / predominant symbol setting unit for inputting a past encoding state from the prediction bit length calculation unit and setting a prediction bit length and a superior symbol of an input bit sequence to be newly inputted; The entry output control signal generator, when receiving the completion signal from the predicted bit length calculator, indicates the number indicated by the new predicted bit length newly set by the predicted bit length / calculated symbol setting unit for the buffer register. A data encoding apparatus for a multi-valued information source, which instructs to take in an input bit sequence. 14. A multi-valued information source consisting of a plurality of bits is divided into bit planes, and encoded data of each bit plane is input and decoded into a binary bit string consisting of "0" and "1". Then, in the multi-level information source data decoding method for decoding the multi-level information source, one of "0" and "1" of each bit plane is set as a superior symbol and the other is set as an inferior symbol. And an input step of inputting, one bit at a time, a codeword representing a prediction result of predicting that the dominant symbol is continuous n (n is an integer of 1 or more) as a binary bit string consisting of “0” and “1”. When the input codeword is a value per prediction, the above-mentioned n predominant symbols are consecutively decoded, and when the number of per-prediction is continuous a predetermined number of times, the number of more than n predominant symbols is consecutive. Then, a data decoding method for a multi-valued information source is newly predicted. 15. A multi-valued information source comprising a plurality of bits is divided into bit planes, and one of “0” and “1” of each bit plane is set as a superior symbol and the other is set as an inferior symbol. And an input step of inputting, one bit at a time, a codeword representing a prediction result of predicting that the dominant symbol is continuous n (n is an integer of 1 or more) as a binary bit string consisting of “0” and “1”. And when the input code word is a value per prediction, decodes the n dominant symbols consecutively, and when the input code word is a non-prediction value, predicts the next code word. A result decoding step and a step of consecutively decoding nm the dominant symbols when the value of the next codeword is a value per prediction, and inputting the next codeword again when the value of the prediction error is obtained. Recursively, A data decoding method for a multi-valued information source, characterized by decoding the inferior symbol at the time of misprediction when 0 <nm ≦ 1. 16. A multi-valued information source consisting of a plurality of bits is divided into bit planes, and code bits serving as coded data of each bit plane are inputted, and a binary value consisting of "0" and "1" is inputted. In the data decoding device of the multi-valued information source for decoding the multi-valued information source after converting the decoded bits into the decoded bits composed of the following bit strings, one of “0” and “1” of each bit plane is set as the dominant symbol, When one of them is the inferior symbol, it receives a superior symbol of the code bit and a prediction bit length setting operation unit for setting n prediction bit lengths, and receives a decoded bit output permission signal from the prediction bit length etc operation unit. A decoding bit setting unit for temporarily holding the input code bit in a predetermined form and outputting a decoded bit, and when the input code bit is the above dominant symbol, Output an output enable signal and write the superior symbol to the decoded bit setting unit,
A data decoding apparatus for a multi-valued information source, wherein the prediction bit length is changed to a number greater than n when the dominant symbol continues for a predetermined number of times. 17. A multi-valued information source composed of a plurality of bits is divided into bit planes, one of “0” and “1” of each bit plane is set as a superior symbol, and the other is set as an inferior symbol. Decoding for decoding by inputting, one bit at a time, code bits that represent the result of predicting that the symbol is continuous (n is an integer of 1 or more) as a binary bit string consisting of “0” and “1”. In the apparatus, a prediction bit length and a superior symbol setting unit for setting the prediction bit length of the code bit and the superior symbol, and a prediction bit length and a superior symbol and the code bit from the prediction bit length and superior symbol setting unit are input. A prediction bit length calculation unit for outputting a decoded bit output permission signal in accordance with the value of the code bit, and the decoded bit output permission signal, A decoding bit setting unit that temporarily holds the current code bit in a predetermined form and outputs a decoded bit, wherein when the code word input to the prediction bit length calculation unit is a value per prediction, the superior symbol Are successively written to the decoding bit setting unit, and when the input code word is mispredicted, the next code bit is input. When the value is a value per prediction, the dominant symbol is set to n− m pieces (m is an integer of 1 or more and smaller than n) are continuously written to the above-mentioned decoded bit setting section, and when the prediction is incorrect, the next code is again inputted to the prediction bit length calculating section. A data decoding device for a multi-valued information source. 18. A multi-level information source composed of a plurality of bits is divided into level planes, and when a binary bit string composed of “0” and “1” of each level plane is input,
A prediction setting step of setting one of “0” and “1” as a dominant symbol, setting the other as a dominant symbol, predicting that the dominant symbol is continuous n, and setting the n as the number of bits to be predicted And when a prediction is hit for the input sequence of interest consisting of the number of prediction bits, one of "0" and "1" is output as a codeword as a per-prediction signal, and the next n bit strings are output. And a prediction result output step of outputting the other signal of either “0” or “1” as a codeword as a mispredicted signal when the prediction is deviated. Data coding of a multi-valued information source characterized by repeating the same prediction setting step and prediction output step recursively with the number of predicted bits being a new reduced predicted number of bits less than n Law. 19. When dividing a multi-valued information source consisting of a plurality of bits into level planes and inputting a binary bit string consisting of "0" and "1" of each level plane,
A prediction setting step of setting one of “0” and “1” as a dominant symbol, setting the other as a dominant symbol, predicting that the dominant symbol is continuous n, and setting the n as the number of bits to be predicted And, when prediction is performed on the input sequence of interest consisting of n bit strings, one of "0" and "1" is output as a codeword as a per-prediction signal, and the next n bit strings are output. And a prediction result output step of outputting the other signal of either “0” or “1” as a codeword as a de-prediction signal when the prediction is deviated. A data encoding method for a multi-valued information source, wherein the same prediction setting step and prediction result output step are repeatedly performed with the number of predicted bits being a new increased predicted bit number larger than n. . 20. A multi-valued information source composed of a plurality of bits is divided into level planes, and among the level planes,
Level planes with high probabilities are treated independently, and low-probability ones are grouped into multiple level planes, and it is determined whether the input multi-valued information source corresponds to each group number The bits are composed of bits consisting of "0" and "1". When this determination bit string is input, the input step is performed from the group number side where the probability is concentrated, and the determination bit "0" or "1" of each group is input. Is set as the superior symbol, and the other is set as the inferior symbol, and the superior symbol is n
A prediction setting step of predicting the number of consecutive bits and setting the number n as the number of bits to be predicted; One of the signals is output as a signal per prediction, and the operation shifts to the operation of encoding the next n bit strings. When the signal deviates, the other signal of "0" or "1" is de-predicted as a codeword. A prediction result output step of outputting as a signal, and a similar prediction setting step and a prediction output step of recursively repeating the same prediction setting step and the prediction output step as a new reduced prediction bit number smaller than n when the prediction is deviated a predetermined number of times. And if the determination bit is a signal indicating that the input signal belongs to the group, the encoding of the input symbol is completed; otherwise, the probability is more concentrated than the group. To determine bit have groups, the prediction set step and the prediction result output process and the data encoding method of the multi-level information source, characterized in that so as to perform the above recursive steps. 21. The above-mentioned n number is 2 ^m (m is an integer of 0 or more).
21. The data encoding method for a multilevel information source according to claim 18, 19, or 20. 22. When the prediction is deviated, the noted sequence is divided into two, and when all the divided first-half first-half target sequences are the dominant symbols, “0” is output as a codeword, and The second half attention series of the second half of the
A first half hitting step of dividing and outputting a code word of “0” or “1”; and outputting a “1” as a code word when the inferior symbol exists in the first half attention sequence, The target sequence is further divided into two, and a first half off step of outputting a code word of “0” or “1” is performed. As long as the inferior symbol exists in each divided target sequence, the division of the target sequence is performed. 22. The data encoding method for a multi-valued information source according to claim 21, wherein the first half contact step and the first half separation step are recursively repeated. 23. A multi-valued information source comprising a plurality of bits is divided into level planes, and among the level planes,
Level planes with high probabilities are treated independently, and low-probability ones are grouped into multiple level planes, and it is determined whether the input multi-valued information source corresponds to each group number Bits are composed of bits consisting of "0" and "1", and either "0" or "1" of the determination bit is set as a superior symbol, and the other is set as an inferior symbol, and the superior symbol is n. A prediction bit length calculation setting unit that predicts that the number of bits is continuous and sets n as the number of prediction bits; a buffer register that temporarily stores an input bit sequence; When a value is input and a prediction is successful for a target sequence consisting of the input prediction bit number, either "0" or "1" is used as a codeword. One of the signal outputs as a prediction hit signal or Re, as a code word when off "0"
Or a determination unit that outputs the other signal of “1” as an out-of-prediction signal, and when the prediction deviates by a predetermined number of times, sets the predicted bit number smaller than n to the new reduced predicted bit number Set in the length calculation setting section,
A data encoding device for a multi-valued information source, wherein when the number of predictions reaches a prescribed number of times, the number of predicted bits is set to a new increased predicted bit number greater than n by the predicted bit length etc. operation setting unit. . 24. A multi-valued information source composed of a plurality of bits is divided into level planes, and among the level planes,
Level planes with high probabilities are treated independently, and low-probability ones are grouped into multiple level planes, and it is determined whether the input multi-valued information source corresponds to each group number The bits are composed of bits consisting of "0" and "1", and hold a buffer register consisting of a group of registers for temporarily storing an input bit sequence consisting of the determination bits, and a variable ofs indicating the head position of the target sequence to be encoded. Bit length calculation unit including a register to be executed and a register holding a variable width indicating the prediction bit length, a position of ofs on the buffer register output by the prediction bit length calculation unit, and its off
s, selects an input bit sequence on the buffer register determined by the width, outputs one dominant symbol as a code bit output when all of the input bit sequences are dominant symbols, and outputs one dominant symbol when a dominant symbol is included. A determination unit that outputs a symbol, an input / output control signal generation unit that issues an input bit request when receiving a completion signal indicating that encoding of the input bit sequence in the buffer register has been completed from the prediction bit length calculation unit, A stack memory serving as a memory for holding the value of the width that changes every moment, and a prediction bit length of an input bit sequence newly input by inputting a past encoding state from the prediction bit length calculation unit And a predictive bit length and dominant symbol setting unit for setting a dominant symbol. When the signal is received from the prediction bit length calculation unit, the buffer register is instructed to fetch the number of input bit sequences indicated by the new prediction bit length newly set by the prediction bit length / operation symbol setting unit. A data encoding device for a multi-valued information source.