JP2794842B2 - Encoding method and decoding method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a highly efficient encoding method for encoding a video signal, an audio signal, and the like at a smaller bit rate for digital transmission and recording, and a method thereof. The present invention relates to a decoding method for performing an inverse transform.

2. Description of the Related Art Various methods have been proposed as highly efficient encoding methods. Among them, the differential pulse code modulation (hereinafter abbreviated as DPCM) method has features such as a relatively simple circuit configuration and good quality of a reproduced signal, although information compression efficiency is low.

Recently, Ramamoorthy et al. Developed a new encoding method to replace DPCM, called Modulo-PCM (hereinafter abbreviated as MPCM) for voice.
A method has been proposed in the following documents [1] to [3]. A new decoder for MPCM has been proposed by Hagiwara et al. In reference [4].

(Literature) [1] Tea. Ericsson and Vi. Ramamusie, “Modulo PCM: A New Source
Coding schemes, "Inconf., Int. Conf. Acoustic., Speech, Signal Processing, ICSA '79, Washington, DC, 1979, P.P. 419-422. (T. Ericson and V. Ramamoorthy," Modelo-PCM: A new
source coding scheme, "in Conf., Int.Conf.Acoust., Sp
eech, Signal processing, ICASSP'79, Washington, DC, 197
9, pp 419-422.) [2] V. Ramamusie, "Speech Coding Using Modulo PCM with Side Information," Incomf. Lek., Int.
Conf. Acousto., Speech, Signal Processing, ICSP'81, Atlanta, GA, Mar, 1981. P. 832-835.
with side information, "in Conf.Rec., Int.Conf.Acous
t., Speech, Signal Processing, ICASSP'81, Atlanta, GA, M
ar.1981, pp.832-835.) [3] V. Ramamusie, "A Nobel Speech Coder for Medium and High Bit Rate Applications Using Modulo PCM Principles," IEE Trans. Acousto., Speech, Signal Processing, Vol. ASP33, P.356-3
68, APR. 1985. (V. Ramamoorthy, “A Novel Speech Coder for Medium
and High Bit Rate Applications Using Modulo-PCM P
rinciples, "IEEE Trans.Acoust ,, Speech, Signal Proces
sing, vol. ASSP33, pp. 356-368, Apr. 1985) [4] Hagiwara and Nakagawa, "New Decoder for Modulo-PCM", 8th Symposium on Information Theory and Its Applications,
(Nara, japan, Dec., 5-7.1985, pp 517-522) The MPCM system has almost the same circuit configuration as the DPCM system, and the presence of granular noise is the same as that of the DPCM, but there is overload noise. Has the advantage of not.

First, the MPCM method will be briefly described. FIG. 9 shows a basic system of the MPCM system.

The MPCM encoding device includes a remainder operation unit 701 and a quantizer 702.

Here the k th sample of the input signal and x _K. The remainder arithmetic unit has a saw-tooth waveform input / output characteristic with an amplitude d as shown in the figure. And the output ((x _K))
Can be expressed as ((x _K )) = x _k −d · [[x _K ]]. Here, ((x _K )) is an arbitrary constant satisfying the following equation (A).

(−d / 2) ≦ ((x _K )) <(d / 2) (A) The output ((x _K )) of the remainder arithmetic unit is quantized into L bits to become a transmission signal y _K , which is decoded. Transmitted to the conversion device.

As described above, in the MPCM encoding apparatus, the respective waveform diagrams are shown in FIG. 10, and as shown in FIG. 10B, the input speech signal is packed into (−d / 2, d / 2) (“packing”). ") To perform information compression. This is the PCM method,
This is equivalent to transmitting only the lower bits.

In the decoding device, as shown in FIG. 9, a first-order predictor sets _K = ρ · _K (ρ is a prediction coefficient) (B), and k samples from the value _{K at} (k−1) sample points predict the value x _K of points. Then, the value of _K is quantized by the characteristic as shown in FIG. 11 and multiplied by d to obtain the upper bit component d. [[X _K ]] which is not transmitted, and this is referred to as the transmission signal y _K
In addition, the input signal of the encoding device is reproduced.

Problems to be Solved by the Invention Next, anomalous errors, which is a major problem of the MPCM system, will be described.

When the absolute value of the difference value Δx _K = x _K −x _k−1 of the input signal becomes larger than d / 2, the upper bit component d ·
[[X _K ]] is not correctly predicted. FIG. 12 shows a waveform chart in that case. Therefore, the decoded output signal _K includes
As shown in FIG. 12 (c), there are many points where the waveform changes sharply. This is the MP called the anomalous error
This is an error peculiar to the CM system, and how to reduce it is
This is an issue for the MPCM system.

Therefore, Ramamoorthy et al. Have proposed two methods for minimizing the occurrence of irregular errors, which are described below (references [2] and [3]).

Method 1: In this method, a decoding device is provided in an encoding device, and when an anomalous error occurs in the decoding device, this is detected and information for correcting the anomalous error is transmitted as side information on another channel. is there.

Method 2: Since the occurrence frequency of the irregular error is high when the prediction error is large, the input data is grouped every N samples to form a block, and the maximum value E of the absolute value of the prediction error is set for each block.
_mN is obtained, and the modulo width of the remainder arithmetic unit and the quantization step of the quantizer are controlled accordingly. Actually the maximum
E _mN is quantized and multiplied by a predetermined coefficient to obtain modulo amplitude d _mN
And transmits this _dmN to the decoding device as side information.

However, the above two methods have the following problems.

In the case of the method 1, since it is uncertain which sample of the encoded input data has an irregular error, it is necessary to send correction information for each input sample in order to completely eliminate the irregular error. This is equivalent to increasing the number of quantization bits, and lowers the coding efficiency. When the number of quantization bits of the quantizer is large, the frequency of occurrence of anomalous errors is small, so the bit rate of the channel for transmitting side information can be lower than that for transmitting information for correcting each sample, but in this case, it is completely possible. Anomalous errors cannot be eliminated.

In the case of the method 2, the occurrence frequency of the irregular error can be reduced by optimizing the various parameters. However, the irregular error cannot be fundamentally removed due to the influence of the quantization error by the quantizer, so the irregular error is not eliminated. Cannot be completely eliminated.

Hagiwara et al. Have proposed a decoding device that corrects anomalous errors by providing a new determination circuit on the decoding device side without sending side information. However, since all of the irregular errors cannot be detected, the irregular errors cannot be completely eliminated. (Reference [4]).

Further, when the input of the MPCM method is an image signal, the irregular error causes visually noticeable deterioration in image quality, and therefore, application to the MPCM method image signal has not been considered.

Means for Solving the Problems The present invention uses a block composed of a plurality of input data as a unit of encoding, and calculates a maximum value SX and a minimum value SN of a prediction error which is a difference between input data and a prediction value of the block. And when a predetermined value larger than (SX-SN) is set as the divisor data OU, each input data of the block is converted to the divisor data.
The remainder E obtained by dividing by OU is encoded, and the SX, SN, OU
And transmitting the coded E together with an additional code including prediction error information related to the coding method.

Further, the present invention provides the coding method, wherein a signal obtained by converting the original input signal so that a dynamic range of a prediction error of the original input signal is equal to or less than a predetermined amount for each of a predetermined number of blocks is used as an input signal. Method.

Further, the present invention uses a block composed of a plurality of input data as a unit of coding, and calculates a maximum value SX and a minimum value SN of a prediction error which is a difference between the input data in the block and its predicted value.
And calculate a predetermined value larger than (SX-SN) as the divisor data.
OU, encode the remainder E obtained by dividing each input data of the block by the divisor data OU,
A signal encoded by an encoding method for transmitting both the additional code including prediction error information regarding X, SN, and OU and the encoded E is input, and the divisor data OU is obtained from the transmitted additional code. , The transmitted decoded remainder E
Is added to the divisor data OU of an integer N times as decoded data, so that a prediction error of a difference between the decoded data and its predicted value satisfies a prediction error range obtained from the transmitted additional code. A decoding method characterized by determining the integer N.

Further, the present invention is a decoding method characterized in that when the decoded data is obtained by converting the original input signal in the decoding method, the original input signal is obtained by performing an inverse conversion.

In the present invention, lossless encoding can be performed by the above configuration, so that irregular errors can be completely eliminated. In addition, when performing encoding at a constant rate, lossless encoding is performed after performing conversion such that the dynamic range of the prediction error is equal to or less than a predetermined amount with respect to the original input signal.
In addition, since the divisor data is determined by the dynamic range of the prediction error, the coding efficiency can be further improved particularly in a monotonically increasing or monotonically decreasing block.

Embodiment Before describing embodiments of the present invention in detail, the principle thereof will be described.

Multiple input data (D1, D2, ……, Dk) with close sample positions
When a block is configured using as a unit of coding, if k input data in the block are taken out one by one and the difference is taken, the difference value is small due to signal correlation, and the dynamic range of the difference value ( Difference between maximum and minimum)
Most of the blocks are smaller than the dynamic range of the input data. Since this difference value is a prediction error when the previous value prediction is performed, it is hereinafter referred to as a prediction error Si. The present invention intends to perform more efficient encoding than the conventional technique, based on the property that most of the blocks whose dynamic range of the prediction error Si is smaller than the dynamic range of the input data.

When the maximum value, the minimum value, and the dynamic range of the prediction error Si in the block are SX, SN, and SDR, the following expression is satisfied: SDR = SX−SN (1) The following formula OU> SDR (2) is satisfied, and a predetermined value uniquely defined by the dynamic range SDR is defined as the divisor data OU.

The quotient and remainder obtained by dividing the input data Di (i = 1, 2,..., K) in the block by the divisor data OU are Ni and E, respectively.
When i is given, the following equation is satisfied: Di = Ei + OU · Ni (3) O ≦ Ei <OU (where Ni is an integer).

By transforming equation (3), the following equation is obtained: Ei = Di−OU · Ni (4) The remainder Ei can be regarded as the input data Di with the offset OU · Ni removed. Let the offset be denoted by Fi. That is, the offset Fi is defined by the following equation: Fi = OU · Ni (5)

Since the dynamic range SDR of the prediction error is mostly smaller than the dynamic range of the input data, if the divisor data OU is set smaller in Equation (2), the divisor data OU is smaller than the dynamic range of the input data. Small blocks can be made most. Therefore, if the input data Di can be restored from the residual data Ei, higher efficiency encoding can be performed by encoding and transmitting the residual data Ei instead of the input data Di. The encoding method of the present invention achieves this.

Hereinafter, the input data Di is obtained from the remainder data Ei. That is, a decoding method will be described.

In order to return the remainder data Ei to the input data Di, it is necessary to reproduce the offset value Fi = OU · Ni. Input data
Assuming that the predicted value for Di is Pi, the maximum value and the minimum value of the prediction error are SX and SN, so the following expression Pi + SN ≦ Di ≦ Pi + SX (6) holds, and the input data Di is expressed by Expression (3). Since this can be expressed, this is substituted into equation (6) to obtain the following equation: Pi + SN−Ei ≦ OU · Ni ≦ Pi + SX−Ei (7) Since the predicted value Pi (i = 2, 3,...) Can be sequentially obtained from the input data obtained by decoding, P1 or F1 (= OU · N1) or N1 is known in addition to SX, SN, and OU. I just need. Since the previous value prediction is performed, the prediction value P1 can be data in an adjacent block that has already been decoded,
It is not always necessary to transmit. Therefore, if the encoding apparatus transmits SX, SN, and OU, which are information on a prediction error for each block, together with the residual data Ei, the offset Fi = OU · Ni can be uniquely determined by equation (7), and the residual data Ei is added. Thus, the data Di is obtained, and the decoding is completed. Furthermore, if any one of P1, F1, N1 is added and transmitted for each block, input data of other blocks is not used, so
Encoding and decoding can be performed independently for each block, and transmission errors can be improved.

The offset Fi is uniquely determined by the definition of OU (Pi
This is because the difference between (+ Sn-Ei) and (Pi + SX-Ei) is smaller than OU. Therefore, OU · Ni satisfying OU · Ni ≧ (Pi + SN−Ei)> OU · (Ni−1) (8) is also an offset Fi, and OU · Ni ≦ (Pi + SX−Ei) <OU · (Ni + 1) OU / Ni that satisfies (9) is also offset Fi. That is, the same offset Fi can be obtained by using any of the equations (7), (8), and (9).

Since the remainder data Ei is a remainder obtained by dividing the input data Di by the divisor data OU, it is 0 or more and less than the divisor data OU. Therefore, the word length of the remainder data Ei is determined by the size of the divisor data OU, and is variable in block units.

The number of bits required to represent an OU is log ₂ OU,
Is smaller, but since the word length L of the remainder data is an integer, a value less than or equal to 2 ^M that satisfies the following expression 2 ^M > SDR ≧ 2 ^{M -1} (10) may be used as the divisor data OU. Word length L
Does not change and the coding efficiency is not improved. Rather, by using ^2M as the divisor data OU, various advantages can be obtained such that the number of bits required to represent the OU can be reduced, and a circuit (for example, a divider) for obtaining an offset can be simplified. This will be described later.

It was stated that SX, SN, and OU are necessary when determining the offset value Fi required to obtain the data Di from the residual data Ei in the decoding device. Are transmitted together with the remainder data Ei, and the offset values Fi required for decoding can be similarly determined using these. Because the following formula Pi + SNN ≤ Di ≤ Pi + SXX (12) (However, the left equal sign is when SNN = SN, and the right equal sign is SXX = SX
Holds when ) Holds.

Since SX, SN, SDR, OU, SXX, and SNN are information relating to prediction errors, they are collectively referred to as prediction error information. SX, SN, OU or
As one of the three data combinations of SXX, SNN, and OU can be obtained from the remaining two data, any two of the three data combinations can be directly or indirectly transmitted as prediction error information. I just need. Further, since the divisor data OU is uniquely determined from the dynamic range SDR of the prediction error, SDR (= SX−SN) may be transmitted as one of the prediction error information instead of the divisor data OU.

In the above description, previous value prediction is used for prediction, but more advanced prediction may be used.

According to the above-described encoding method, the prediction error information (SX, SN, OU) and the residual data obtained by encoding are obtained by dividing the divisor data from all the input data Di in the block.
It becomes equal to prediction error information obtained by encoding data obtained by removing a constant value that is an integral multiple of OU and remainder data. This is because the prediction error is the same even if a certain value is removed from all the input data, and if the certain value is an integer multiple of the divisor data OU, then the remainder data, which is the remainder divided by the OU, is the same. Therefore, the following decoding method is also possible. That is, if there are all the residual data Ei (i = 1, 2,..., K) and prediction error information (for example, SX and OU) in the block, the input of the initial value F1 (or P1, or N1) Even if the decoding is not completed, the decoding can proceed with F1 = 0, and the result is obtained (Di-F1).
Then, the obtained initial value F1 is added to the decoded data Di.

FIG. 1 is a block diagram showing a configuration of a high-efficiency encoding device according to the first embodiment of the present invention.

In FIG. 1, reference numeral 101 denotes an input terminal of input data which is a sampled and quantized video signal, and 102 divides a screen represented by the video signal into small blocks, and inputs data Di (i = 1, 2, ……, 16) Block divider that outputs 1)
03 is a delay unit for adjusting the timing of input data for one block for timing adjustment, 104 is a register for holding the leading data D1 of each block output from the delay unit 103, and 105 is an output from the block divider 102. A predictor that includes a register and outputs the previous input data as a predicted value Pi;
106, a subtractor for subtracting the predicted value Pi from the output of the block divider 102 to obtain a prediction error Si (i = 2, 3,..., 16);
Is the maximum value SX of the prediction error Si from the subtractor 106 for each block.
And a minimum value detector 108 for obtaining a minimum value SN and holding and outputting the result in an internal register for each block. 108 subtracts the minimum value SN from the maximum value SX to calculate the dynamic range of the prediction error.
A subtractor that outputs SDR, 109 is the dynamic range SD
A divisor data generator 110 which receives R and outputs divisor data OU, 110 divides the input data Di from the delay unit 103 by the divisor data OU, and generates remainder data Ei (i = 2, 3,..., 16)
, 111 is a remainder encoder that encodes the remainder data Ei according to the size of the divisor data OU to obtain encoded data Ci, 112 is an output terminal of the encoded data Ci, and 113 is the divisor data OU. Is an output terminal for the maximum value SX of the prediction error, and 115 is an output terminal for the leading data D1 in the block.

The operation of the encoding apparatus of the present embodiment configured as described above will be described below.

Input data (a word length of 8 bits at an amplitude level of 0 to 255), which is a sampled and quantized video signal, is input from a terminal 101 to a block divider 102. The block dividing circuit 102 includes a buffer memory therein, and sequentially writes input data in the buffer memory and simultaneously inputs data Di in blocks.
(I = 1, 2,..., 16) is read out. Input data D1 output first from each block is held by the register 104 and output from the terminal 115 as one of the additional codes.
The input data input to the predictor 105 is output as a predicted value Pi and supplied to the subtractor 106. The prediction value Pi is subtracted from the input data Di in the subtractor 106 to obtain a prediction error Si. In the maximum / minimum detector 107, the prediction error S
i is input, and the maximum value SX and the minimum value SN of the prediction error Si in the block are output. The minimum value SN is subtracted from the maximum value SX by the subtractor 108 to obtain the dynamic range SDR of the prediction error. The dynamic range SDR is input to the divisor data generator 109, and the divisor data OU that satisfies the following equation (OU = ^2M > SDR ≧ ^2M-1 ) is output. The input data Di whose timing has been adjusted by the delay unit 103 is output to the remainder arithmetic unit 11.
At 0, the remainder data Ei divided by the divisor data OU
(I = 2, 3,..., 16) and output. Since the OU is ^2M , all of the remainder data Ei that is less than the OU can be represented without deterioration with a word length of M bits. Since the remainder data Ei does not become larger than the input data Di, its word length is 8 or less. The remainder data Ei input to the remainder encoder 111 is encoded into a bit number J (where J = M when M <9 and J = 8 when M = 9) determined by the size of the divisor data OU, and has a variable length. The encoded data Ci is output from the terminal 112. Residual data E during encoding
Since non-linear processing such as truncation of lower bits is not performed on i, the residual data Ei does not deteriorate due to encoding. That is, lossless encoding is performed. Therefore, the residual data E obtained by decoding the encoded data Ci
i ′ matches the original remainder data Ei. The divisor data OU as the prediction error information and the maximum value SX of the prediction error are
It is output from 3,114 as one of the additional codes.

FIG. 2 shows an example of input data Di, offset Fi, prediction error Si, and residual data Ei when the block configuration is 16 pixels in the horizontal direction and 1 line in the vertical direction. The 16 input data in this block are 112, 105, 113, 12
1,136,160,175,201,207,208,199,193,176,169,178,190
It is. In such a one-dimensional block configuration, the block divider 102 does not require an internal buffer memory,
Since the predictor 105 only needs to have one register for holding the previous input data, the configuration of the encoding apparatus is simpler. The order in which data is extracted from the two-dimensional block,
The prediction method will be described later.

The maximum value and the minimum value of the maximum value SX of the prediction error that can occur are 255 and −255, respectively. Similarly, the maximum value and the minimum value of the minimum value SN of the prediction error are 255 and −255, respectively. The word length of the maximum value SX and the minimum value SN requires 9 bits. Since the maximum value and the minimum value of the dynamic range SDR of the prediction error are 510 and 0, respectively, the dynamic range SDR
Requires a word length of 9 bits. Accordingly, the word length of the divisor data OU generally requires 9 bits, but in the present embodiment, the divisor data OU is set to be a power of 2 that satisfies the expression (10), so that less bits are required. It can be represented by a number. That is, instead of directly transmitting the divisor data OU, the exponent M in the power-of-two representation may be transmitted. The exponent M has 10 values of 0, 1, 2,..., 9, and its word length may be 4 bits. That is, the divisor data OU is 4
It is transmitted by the bit code M.

Since the divisor data OU is uniquely determined by the dynamic range SDR of the prediction error, the divisor data generator 109 can be composed of a 9-bit input, 4-bit output ROM (read only memory). But the divisor data OU is 2
, The divisor data generator 109 is provided by a simple logic circuit for detecting the number of consecutive 0s from the most significant bit in the data representing the dynamic range SDR.
Can be configured.

Divide the input data by the divisor data OU to obtain the remainder data
The remainder arithmetic unit 110 for obtaining Ei can also generally be configured by a divider or a ROM. However, in the present embodiment, the divisor, OU, is a power of 2, so
A simple logic circuit that extracts only the lower J bits of the input data Di (where J = M when M <9, and J = 8 when M = 9), and sets the other bits to 0 is obtained from the remainder Ei divided by OU. Thus, the remainder arithmetic unit 110 can be realized.

A remainder encoder 111 for encoding the remainder data Ei to the bit number J determined by the size of the divisor data OU
Can be realized by a simple circuit which extracts only the lower J bits of the remainder data Ei and does not output unnecessary upper bits. Although there are various output forms of the encoded data Ci, in this embodiment, it is assumed that the encoded data Ci is outputted in the form of serial data.

Although not shown in the figure, the additional codes (D1, SX, OU) and the encoded data (C2, C3,.
16) is temporarily stored in a buffer memory, subjected to encoding processing for error correction, etc., and then output to a transmission path in the form of, for example, serial data signals.

In the encoding apparatus of this embodiment, the total number of bits of input data Di (i = 1, 2,..., 16) per block is 128 (= 8).
× 16) bits, and the total number of output bits per block CD is 21 (= 8 +) of the additional code (D1, SX, OU) of fixed word length
9 + 4) bits and variable length coded data (C2, C3, ..., C
This is the sum of 16) and 15 · J bits, and can be expressed by the following equation: CD = 21 + 15 · J (13) Although the word length J of the coded data Ci is 0 or more and 8 or less, since the prediction error is small and the word length J is 7 or less due to the correlation of the input signal in most of the blocks, the following equation CD = 21 + 15 · J <128 holds, and the average number of bits per sample (one pixel) can be reduced. When the word length J is 8, 9, the encoding efficiency is deteriorated. Therefore, it is better to transmit the input data as it is. FIG. 2 shows an example of each data in a block having a one-dimensional configuration. According to FIG. 2, the divisor data OU is 64 and the word length J of the encoded data Ci is 6. Therefore, the total number of output bits CD of the encoder in this block is 111 bits (<128) according to equation (13), and the average number of bits per sample is reduced to about 6.9 bits.

FIG. 3 is a block diagram showing a configuration of a decoding apparatus according to the first embodiment of the present invention. This decoding apparatus performs an inverse transform of the encoding apparatus shown in FIG.

In FIG. 3, reference numeral 201 denotes an input terminal of coded data Ci, 202 an input terminal of the divisor data OU, 203 an input terminal of the maximum value SX of the prediction error, 204 an input terminal of the first data D1 in the block, 205 Performs the inverse conversion of the remainder encoder 111 in the high-efficiency encoder shown in FIG. 1. The remainder decoder 206 decodes the coded data Ci to obtain the remainder data Ei '. 206 denotes the prediction error. 207 is a subtractor that subtracts the remainder data Ei ′ from the output of the adder 206, and 208 is an input that receives the output of the subtractor 207 and the divisor data OU. An offset regenerator 209 for outputting the offset Fi ', and the decoded data D that has been decoded by adding the remainder data Ei' and the offset Fi '.
i is an offset adder for obtaining i ', 210 is a switch which receives the output of the offset adder 209 and the data D1 from the terminal 204 and selects one of them, and 211 is the same as the predictor 105 in FIG. ′ As an input and the predicted value Pi
A predictor that supplies the other input of 206, 212 is the switch
A block decomposer 213 which receives the decoded data Di 'from 210 as input and performs the reverse processing of the block dividing circuit 102 in FIG.
Is an output terminal for the decoded data. It is assumed that the input of the additional code precedes the input of the encoded data and is held for one block period.

The operation of the decoding device configured as described above will be described.

The input data D1 from the terminal 204 is output as it is as decoded data D1 'through the switch 210 and supplied to the block decomposer 212 and the predictor 211. The data D1 'becomes a predicted value P2' necessary for obtaining decoded data from the first remainder data E2 'in the predictor 211. Switch 21
0 outputs the input data D1, selects the input from the offset adder 209, and decodes the data D2 ', D3',..., D16 '.
Is output.

The coded data Ci (i = 2, 3,..., 16) from the terminal 201 is input to the remainder decoder 205 based on the divisor data OU from the terminal 202.
Is converted to remainder data Ei ′. The maximum value SX of the prediction error from the terminal 203 and the prediction value Pi 'from the predictor 211 are added by the adder 206, and the remainder data Ei' is subtracted by the subtractor 207 from the addition result (Pi '+ SX). −E
i ′) is obtained. The subtraction result and the divisor data OU are input to the offset regenerator 208, and the offset Fi ′ = OU · that satisfies the following equation: OU · Ni ≦ (Pi ′ + SX−Ei ′) <OU · (Ni + 1) (9) Ni is obtained. In the offset adder 209, the offset Fi 'and the remainder data Ei' are added, and the decoded data Di '(i = 2, 3,...)
, 16) and supplied to the switch 210.

At the output of switch 210, the decoded data D in the block
All of i ′ (i = 1, 2,..., 16) are obtained and supplied to the predictor 211 and the block decomposer 212. The decoded data Di 'input to the block decomposer 212 is extracted in the same order as data obtained by sampling and quantizing the original video signal, and
Output from 213.

Since the word length of the encoded data Ci input from the terminal 201 in the form of serial data is variable in block units, it is necessary to know the word length J to correctly extract the encoded data Ci. Since the word length J of the encoded data Ci can be determined by the index M representing the size of the divisor data OU (however, when M <9, J = M, M = 9 J = 8), the coded data Ci can be accurately extracted, and the remainder data Ei 'can be obtained by adding 0 to its upper bits. Therefore, the remainder decoder 205 can be realized by a simple circuit such as a shift register for serial / parallel conversion, a register for holding encoded data, a counter for counting the number of bits, and a gate for adding 0 to upper bits. Residual encoder 111 in FIG.
Is not degraded by encoding, and the remainder decoder 205 performs the inverse transform of the remainder encoder 111, so that the remainder data Ei 'matches the remainder data Ei.

The output of the subtractor 207, (Pi '+ SX-Ei'), is divided by the divisor data OU to obtain a remainder, and the remainder is calculated as (Pi '
+ SX−Ei ′) to obtain an offset Fi ′ = OU · Ni satisfying the expression (9). However, in the present embodiment, the divisor data OU is set to 2 to the power of M, so that (Pi '+
The offset Fi 'can be obtained simply by setting all lower M bits of SX-Ei') to "0". Therefore, the offset regenerator 20
8 can be realized with a simple logic circuit.

Since the residual data Ei ′ and the predicted value Pi ′ in the decoding device are equal to the residual data Ei and the predicted value Pi in the encoding device of FIG. 1, the offset Fi ′ is also equal to the offset Fi, and the decoded data Di ′ is the input data Di. Matches.

As described above, according to the present embodiment, there is no quantizer such as the MPCM, the residual data is encoded and transmitted as it is, the prediction error change range, the divisor data OU
(Modulo width in the MPCM system) is set to be larger than the prediction error change range (dynamic range), and the divisor data OU and the prediction error change range information are transmitted as additional codes. And lossless encoding, that is, reversible encoding and decoding thereof can be realized, and the circuit configuration is extremely simple.

In the conventional example, the divisor data (mo
The width of the dulo) is fixed or the maximum value of the absolute value of the prediction error. However, in the present invention, the divisor in encoding can be made smaller than that of the conventional example because it is determined by the change range (dynamic range) of the prediction error. Therefore, particularly in a block in which input data monotonically increases or decreases, more efficient encoding can be performed than in the conventional example. When an image signal is input, a coarse image can be reproduced using only a part (additional pixel data of the block) of the additional code, so that encoding suitable for image search can be performed.

An encoding method in which the average value DA of the input data in the block is transmitted as one of the additional codes instead of the input data at a specific position in the block is also conceivable. As a decoding method in this case, first, decoding is performed with the initial value of off-foot F1 = 0, and an average value DB of the obtained data (Di-F1) is obtained. The initial value F1 of the offset is obtained by subtracting the average value DB from the transmitted average value DA. The data (Di-
By adding the obtained initial value F1 to F1), decoded data Di is obtained. If all k pieces of remainder data Ei are transmitted, even if one error occurs in the remainder data, the error can be corrected by using the average value DA. When the input is an image signal, an outline of the image signal can be represented using only the average value DA, and encoding suitable for search can be performed.

Next, a case where an error occurs in encoded data on a transmission path will be described.

Even if a code error occurs on the transmission path, the error does not necessarily propagate to the decoding of the next sample as in the DPCM method.
This is because the determination of the offset Fi by the equations (7), (8), and (9) has a range of the determination determined by the maximum value and the minimum value of the prediction error S. However offset
An error may occur in the determination of Fi, and an error of the offset Fi propagates to decoded data after this time. However, when the prediction value Pi is obtained, data in another block is not used, that is, since encoding is performed independently for each block, an error does not propagate to an adjacent block. Also, since the difference between the correct decoded data and the erroneous decoded data due to the propagation of the error is an integral multiple of the divisor data OU, it is necessary to determine the correlation between the decoded data in which the error has propagated and the correct decoded data adjacent thereto. It is possible to correct with higher accuracy.

The correction method will be described. The position where the error has occurred can be detected by the error correction code. For example, FIG. 2 (c)
It is assumed that a code error has occurred in the ninth residual data shown in FIG. As a result, the decoding of the tenth to sixteenth data is erroneous. Therefore, decoding is performed while predicting the 17th data (not shown in FIG. 2), which is the head data of the next block in which no error propagates, as a starting point in the direction opposite to the encoding. As long as the difference between the 17th data and the 16th data does not exceed the dynamic range of the prediction error, the 16th, 15th,
... Correct decoding can be performed up to the tenth input data. However, since the prediction direction is different from that at the time of encoding, the sign of the prediction error is inverted, and the maximum value and the minimum value used for decoding are obtained by inverting the signs of the minimum value and the maximum value transmitted by the additional code, respectively. There is a need. The ninth input data can be corrected by the average value of the eighth and tenth decoded data.

In addition, a prediction error obtained by performing prediction with a block boundary between decoded data (the 16th and 17th in the example of FIG. 2).
The region in which the difference between the second data and the second data is out of the range of the prediction error obtained from the additional code can be determined to have a high possibility that a transmission error has occurred. If a transmission error can be determined,
Similarly to the above method, an out-of-range area can be predicted and decoded by a method different from that at the time of decoding, using decoded data in an adjacent block as a starting point.

In the case of FIG. 2, the block configuration is one-dimensional. However, in the case of two-dimensional or more like an image signal, it is possible to correct using the decoded data in a plurality of surrounding blocks. When a transmission error occurs in the additional code representing the prediction error information, a correction method using the additional code of the adjacent block can be considered. Further, it is possible to determine at the time of encoding which adjacent block is appropriate for use at the time of correction at the time of transmission error at the time of encoding and to transmit this information as one of the additional codes.
In this case, more appropriate correction can be performed.

The first embodiment is a variable-length coding in which the number of coding bits changes for each block, and is more suitable for applications that do not require real-time transmission, such as recording of files such as still images and audio of short sentences. I have.

Next, the principle of the encoding method of the present invention adapted to a normal transmission path having a constant bit rate will be briefly described.

One method for keeping the bit rate constant is to keep the number of coded bits per block constant. To keep the number of encoded bits constant means to make the dynamic range of the prediction error equal to or less than a predetermined amount. According to the present invention, a conversion input signal is obtained by subjecting the input signal to conversion so that the dynamic range of the prediction error is equal to or less than a predetermined amount, and the reversible encoding of the present invention is performed on the obtained conversion input signal. Things. Further, the encoded output is subjected to the decoding of the present invention to reproduce the converted input signal, and the input signal is reproduced by performing the inverse conversion to that at the time of encoding.

The simplest method for obtaining the converted input signal is to multiply the input signal by a predetermined number for each block.
By multiplying the input signal by a predetermined number smaller than 1, a converted input signal with a reduced dynamic range is obtained, and the dynamic range of the prediction error of the converted input signal is
The dynamic range of the prediction error of the input signal can be reduced by the predetermined number of times. This is because the process of obtaining the prediction error from the input signal is linear.

FIG. 4 shows a block diagram of an encoding device and a decoding device according to a second embodiment of the present invention. This coding apparatus is to make the number of coding bits for each block constant.

4. In FIG. 4, reference numeral 401 denotes an input terminal of input data which is a sampled and quantized video signal (word length of 8 bits 0 to 255 levels), and 402 denotes an encoding device for making the word length of encoded data of each block constant. And 403, a decoding device for performing the inverse transform of the encoding device 402, and 404, an output terminal of the composited input signal.

In the encoding device 402, reference numeral 102 denotes a block divider (same as that in FIG. 1), and 405, 406, 407, and 408 each denote a coefficient 1 /
Coefficient multiplier for multiplying input signal by 2,1 / 4,1 / 8,1 / 16,409,410,41
1, reference numeral 412 denotes an encoding device having an internal configuration in which the block divider 102 is removed in FIG. 1 and performing lossless encoding to output an additional code and encoded data; and 413, encoding devices 409 to 412
A detector 414 which receives the divisor data OU as one of the additional code outputs of the encoding device and outputs the conversion information, and 414 receives the outputs of the encoding devices 409 to 412 and the coefficient unit 408 as inputs and is controlled by the conversion information. If the encoded data word length becomes 4 bits based on the outputs of the encoding devices 409 to 412, the switch selects and outputs the output. Otherwise, the switch selects and outputs the output of the coefficient unit 408.

In the decoding device 403, 212 is a block decomposer (the same as that in FIG. 3), 415 is a decoding device having an internal configuration in which the block decomposer 212 is removed in FIG. 3, and 416 is an output of the decoding device 415. And coded data as input, a switch for controlling the transmitted conversion information to output one of them, a switch 417 for selectively outputting a predetermined coefficient controlled by the transmitted conversion information, and a switch 418 for reproducing from the switch 416. A multiplier that reproduces an input signal by multiplying the converted signal by a predetermined coefficient from the switch 417.

The operation of the encoding apparatus of the present embodiment configured as described above will be described below.

In the encoding device 402, the input signal from the terminal 401 is converted into a signal for each block by the block divider 102. The coefficient units 405 to 408 multiply the input signal of each block by a predetermined coefficient 1/2, 1/4, 1/8, 1/16, and round the fractional part to a word length of 8, 7,
A converted input signal (including an input signal) of 6, 5, or 4 bits is created, and lossless encoding is performed by each of the encoding devices 409 to 412. The detector 413 detects an encoding device in which the word length of the encoded data is 4 bits. Switch 414 is detector 4
13 is controlled by the output of 13 to select and output the output of the encoding device in which the word length of the encoded data is 4 bits. However, the encoding device in which the word length of the encoded data is 4 bits can be detected. If not, the output of the coefficient unit 408 is selected and output as is as encoded data. Since it is necessary to notify the decoding device of information (3 bits) indicating which conversion input signal has been encoded and transmitted, this information is included in the additional code and transmitted as conversion information. It is assumed that encoding apparatus 409 has the lower limit of the encoded data word length set to 4 bits. Therefore, the amount of information to be transmitted is as follows: the additional code has 3 bits of conversion information, 8 bits of block head data (worst value), maximum value of prediction error of 9 bits (worst value), and coded data of 4 × 15 bits, for a total of 80 bits. The average is 5 bits per pixel.

In the decoding device 403, the decoding device 415 decodes the transmitted coded data and decodes the converted input signal in the coding device without error. The decoded converted input signal is input to the multiplier 418 via the switch 416. Switch 417
Selects and outputs an inverse transform coefficient for converting the decoded converted input signal into an input signal under the control of the transmitted conversion information. The multiplier 418 multiplies the inverse transform coefficient from the switch 417 by the transform input signal to convert it into an input signal. The reconstructed input signal for each block is decomposed by a block decomposer 212, and an input signal having the same data sequence as the input signal of the block divider 102 is output from a terminal 404. If the transmitted coded data is detected as an output of the coefficient unit 408 based on the transmitted conversion information, the coded data is the conversion input signal itself.
, And directly to the multiplier 418 via the switch 416.

If the input signal is coded as described above and cannot be coded with a predetermined coded data word length, the input signal is multiplied by a predetermined coefficient to create a converted input signal having a small dynamic range of prediction error, and coded. By doing so, the encoded data word length can be made constant. Since the encoding of the transformed signal is reversible by the same encoding method as in the first embodiment, no irregular error occurs. When the dynamic range of the prediction error of the input signal is large, the input signal is multiplied by a predetermined coefficient, converted into a converted input signal, and coded. The place where the dynamic range of the prediction error is large is where the input signal changes drastically. In a place where the signal changes drastically, signal deterioration is not felt even if the quantization is made coarse, so that it is possible to carry out a practical coding suitable for the visual characteristics.

Although four encoding devices 409 to 412 are shown for ease of explanation, most of the circuits inside the encoding device can be shared, so that the circuit scale is not significantly increased as compared with the first embodiment. .

In the second embodiment, since the conversion coefficients for converting the input signal into the converted input signal are fixed and the types thereof are small, the dynamic range of the encoded data may not be effectively used in some cases. An embodiment of the present invention in which this problem has been solved will be described below.

FIG. 5 shows a block diagram of an encoding apparatus according to the third embodiment of the present invention.

In FIG. 5, reference numeral 501 denotes an input terminal of input data which is a sampled and quantized video signal, 102 denotes a block decomposer (same as 102 in FIG. 1), and 502 generates conversion information and conversion coefficients. A conversion circuit for multiplying the input signal by a conversion coefficient to output a conversion input signal; 503, an encoding device; 504, an output terminal for the conversion information (inverse conversion coefficient 1 / a); output terminal of the data d _1, 506 is an output terminal of the maximum value sx of predictive errors for the conversion input data, is 507 an output terminal for the coded data ci.

In the conversion circuit 502, 508 is a delay unit, 509 is a predictor, 510
Is a subtractor, and 511 is a maximum / minimum detection circuit, which is equivalent to the delay unit 103, predictor 105, subtractor 106, and maximum / minimum detection circuit 107 in FIG. 1, respectively, and 512 is a prediction error from the maximum / minimum detection circuit 511. A conversion information generating circuit for outputting conversion information (conversion coefficient a, inverse conversion coefficient 1 / a) for generating a conversion input signal from the maximum value and the minimum value of the input signal; It is a multiplier that outputs a signal.

In the encoding device 503, the same components as those in FIG. 103 is a delay device, 104 is a register, 10
5 is a predictor, 106 is a subtractor, 110 is a remainder arithmetic unit, 111 is a remainder encoder, 514 detects the maximum value of the prediction error from the subtractor 106 for each block, and holds it in an internal register for each block. This is a maximum value detection circuit to be output.

The operation of the encoding apparatus of the present embodiment configured as described above will be described below.

The input data, which is a sampled and quantized video signal, is
The signal is input from the block 501 and is converted into a signal for each block by the block divider 102. The input signal for each block is multiplied by a conversion coefficient a determined for each block in the conversion circuit 502, and the decimal part is rounded to become a converted input signal. A method for determining the conversion coefficient a will be described later.

The transformed input signal is encoded in the encoding device 503 in the same manner as in the first embodiment of the present invention shown in FIG. At this time, the conversion input signal is converted by the conversion
Are converted so that lossless encoding can be performed with a fixed word length of L bits. Therefore, since it is not necessary to newly obtain the divisor data OU, the encoding device 503 has a configuration in which the subtractor 108 and the offset generator 109 are removed from the encoding device of FIG. Since each signal is a single signal, the block divider 102 is also eliminated. Since the coded data word length L is constant, that is, the divisor data OU is always constant, there is no need to transmit it as an additional code.
The encoding device 503 outputs the maximum value sx of the prediction error of the transformed input signal, the head data d1 of the block, and the encoded data. Further, the conversion information (inverse conversion coefficient 1 / a) from the conversion circuit 502 is transmitted as one of the additional codes because it is necessary for inversely converting the coded and transmitted conversion input signal into an input signal.

The conversion coefficient a is obtained as follows. In the conversion circuit 502, the difference between the output of the predictor 509 and the input signal, that is, the prediction error is obtained by the subtractor 510. The maximum / minimum detection circuit 511 calculates and outputs the maximum value SX and the minimum value SN of the prediction error of each block of the input signal. Thus, the prediction error and the dynamic range SDR of the input signal are (SX-SN). On the other hand, the encoding device 5
Since the word length of the coded data of 03 is L bits, the dynamic range sdr of the prediction error of the converted input signal (when the maximum value of the predicted error of the converted input signal is sx and the minimum value is sn, sdr
= Sx−sn) must satisfy the following expression from Expression (10).

sdr ≦ (2 ^L −1) (14) If there is no rounding error in the multiplier 514, the conversion coefficient a is set to (2 ^L −1) / (SX−SN) to obtain the prediction error of the conversion input signal. Dynamic range sdr (2 ^L −
1) and can be efficiently coded. However, there is actually a rounding error, and sx = R (a · a
SX), sn = R (a.SN) does not hold (correctly, R (x) represents a value obtained by rounding the decimal part of x). Therefore, even if a rounding error occurs, the conversion coefficient a is set to be smaller than (2 ^L -1) / (SX-SN) so that the prediction error sdr always satisfies Expression (14). That is, sx may be larger than R (a · SX), or sn may be smaller than R (a · SN). At this time, sdr becomes larger than a · SDR. Since the maximum value of each error is 1, the conversion information generation circuit 512 generates a conversion coefficient a that satisfies the following equation so that the equation (14) is satisfied even if the influence of the rounding error is in the worst state.

(2 ^L −3) ≧ R (a · SX) −R (a · SN) (15) The conversion information generation circuit 512 previously obtains a conversion coefficient a that satisfies Expression (15) by calculation. This can be easily realized by read-only memory (ROM) in which this is written. The dynamic range SD of the prediction error of the input signal
When R satisfies the following equation: SDR ≦ (2 ^L −1) (16), the conversion coefficient is 1.

FIG. 6 is a block diagram of a decoding device according to a third embodiment of the present invention. The decoding device performs an inverse transform of the encoding device shown in FIG.

In the sixth figure, an input terminal of the conversion information (inverse transform coefficients 1 / a) 601, an input terminal of the distal end data d ₁ in a block of transformed input signal 602, the maximum value of the prediction error for converting input data 603 sx 605 is a decoding device for the transmitted converted input signal, 606 is an inverse conversion circuit for inversely converting the decoded converted input signal to the original input signal, and 212 is a block that has been decomposed and reproduced. Block decomposers 212 and 607, which convert the input signal of each block into input signals having the same data sequence as the input signal of the block divider 102, are output terminals of the reproduced input signal. In the decoder 605, the same components as those in FIG. 205 is the remainder decoder, 2
06 is an adder, 207 is a subtractor, 208 is an offset regenerator, 20
9 is an adder, 210 is a switch. In the inverse transformer 606, 608 is a multiplier.

The operation of the decoding device configured as described above will be described.

The decoding device 605 has basically the same configuration as the decoding device shown in FIG. 3 except that the divisor data OU is not input because it is not necessary to transmit the divisor data OU at a constant value. It becomes the original converted input signal when the encoded data is decoded. The converted input signal becomes the original input signal after being multiplied by the inverse transform coefficient 1 / a (conversion information) transmitted in the inverse transform circuit 606. Since this input signal is a signal for each block, it is converted by the block decomposer 212 into a signal having the same data sequence as the input signal of the encoding device, and output from the terminal 607.

As described above, in the third embodiment shown in FIGS. 5 and 6, a conversion input signal is obtained by multiplying the input signal by the conversion coefficient a, and the configuration of the conversion input signal is shown in FIG. The same reversible encoding as in the first embodiment is performed. Since the dynamic range of the prediction error of the input signal and the dynamic range of the prediction error of the converted input signal are in a proportional relationship except for the influence of the rounding error, first, the dynamic range of the prediction error of the input signal is obtained. Is set so that the dynamic range does not exceed the dynamic range determined by the word length of the encoded data and approaches the dynamic range, encoding can be performed efficiently. In addition, since the transform coefficient is set so as not to exceed the upper limit of the dynamic range in consideration of the rounding error, the transform input signal is reversibly encoded, and no irregular error occurs. The distortion of the input signal due to the encoding and decoding is only due to the word length limitation and the rounding error due to the conversion coefficient multiplication in the conversion from the input signal to the converted input signal and the inverse conversion.

In the above example, the word length of the coded data in each block is all constant, but there is a method in which the average word length of the coded data is constant for each of a predetermined number of blocks. For example, consider a case where the average word length of encoded data is encoded by 4 bits for every two blocks. If the two blocks A and B can be reversibly encoded with 6 bits and 4 bits, respectively, the encoding can be performed without deterioration because the average encoded word length is 4 bits. However, when the post-length of the coded data is fixed to 4 bits for each block, the signal of the block A is deteriorated by 2 bits, and the block has an empty space of 2 bits per sample. Therefore, when the transmission rate is fixed, coding efficiency can be improved by performing coding for each of a plurality of blocks as compared with the case where coding is performed for each block.

In this encoding method, as an example of a method of determining how many bits each of the blocks in the encoding unit are encoded, the following is conceivable. First, the dynamic range of the prediction error is calculated for all the blocks in the coding unit, and the average number of bits of the coded data when lossless coding is performed is obtained. The conversion is performed such that the dynamic range of the prediction error of the input signal of each block is reduced on average by the number of bits corresponding to the difference between the number of bits and the actual number of encoded bits.

This encoding apparatus includes a buffer memory for an encoding unit and a means for obtaining an average number of encoded bits when lossless encoding is performed, and calculates a difference between the average number of encoded bits and the actual number of encoded bits. Means for obtaining and storing the information may be added to the encoding device shown in FIG. 5. Since this decoding device may be basically the same as the decoding device shown in FIG. 6, the description thereof will be omitted.

Further, even in the above embodiment, the dynamic range of the encoded data is not completely used effectively.
This is because a margin is provided so that the dynamic range of the prediction error does not exceed the dynamic range of the encoded data even if the dynamic range of the prediction error is expanded due to the rounding error. So the
2. A method combining the third embodiment is conceivable. That is, in the third embodiment, a plurality of transform coefficients and a plurality of encoding devices corresponding to the transform coefficients are provided, and one that effectively uses the dynamic range of encoded data is selected. Since the configuration can be easily realized by the second and third embodiments, the description is omitted.

As a method for reducing the dynamic range of the prediction error, a method of performing a non-linear conversion as shown in FIG. 7 can be considered. In FIG. 7, white circled data is an input signal before conversion, and black circled data is an input signal after conversion. The example shown in the figure is a case where the maximum value SX (= D _i −D _i-1 ) of the prediction error occurs between the input data at time (i−1) and i, and the input related to the generation of the maximum value. The data _Di-1 and _Di are respectively +1 and -1 to be converted input signals. Thus, the maximum value sx of the prediction error of the converted input signal is two levels smaller than the maximum value SX. As described above, the maximum value and the minimum value of the prediction error generated by slightly or slightly increasing a part or all of the input data related to the generation of the maximum value and the minimum value of the prediction error are slightly reduced. . That is, non-linear processing for reducing the dynamic range of the prediction error is performed. Since the change level of the input data is kept small, the inverse conversion at the time of decoding can be made unnecessary. The conversion by the non-linear processing is effective when the dynamic range of the prediction error of the input signal is equal to or slightly more than a power of two, and the conversion signal is obtained by multiplying the input signal by a conversion coefficient. This is effective in suppressing an increase in prediction error caused by a rounding error. This is because setting the dynamic range of the prediction error to a value smaller than a power of 2 but closer to it means that the dynamic range of the encoded data is effectively used, and the encoded data word length can be reduced by 1 bit. .

Another method for reducing the dynamic range of the prediction error is as follows.

The prediction error can be regarded as a high frequency component of the input signal in the encoding device. Therefore, as a method of reducing the dynamic range of the prediction error, a high-frequency suppression filter can be considered. At the time of decoding, a filter having characteristics opposite to those of the high-frequency suppression filter may be used. In this case, the conversion information may be a code indicating the presence or absence of the filter and its characteristics.

The block may have a one-dimensional configuration (for example, 16 pixels in the horizontal direction, one line in the vertical direction) or a two-dimensional configuration (for example, four pixels in the horizontal direction, four lines in the vertical direction), and a three-dimensional configuration expanded between fields and between frames. Conceivable. As a modification, a method of performing encoding in a field or a frame, and then expanding and encoding a block in a time axis direction in a field unit or a frame unit may be considered.
A method of making the block size variable is also conceivable.

In general, a block is composed of input data at sample positions that are continuous in the horizontal and vertical directions, but is not necessarily limited to such a configuration. A block may be composed of a plurality of input data having a strong correlation. In the case of an NTSC signal or the like, a block may be composed of input data at a sampling position at the same phase point of a chrominance carrier.

In the above embodiments, all the prediction methods are pre-prediction. However, the present invention is not limited to this, and various prediction methods are conceivable.

When the block configuration is one-dimensional, it is easiest to take out input data in the block in the order of input and to combine input data or decoded data used for prediction with the immediately preceding data. When the dimensions are two or more, various combinations of the order of input data extraction and data used for prediction are conceivable. FIG. 8 shows an example in which the block configuration is two-dimensional with four pixels in the horizontal direction and four lines in the vertical direction. In the figure, a circle represents a pixel, a number represents an order i of taking out from a block, and an actual battle arrow represents that data with an arrow is used for predicting data with an arrow. Divisor data
The smaller the OU, the better the coding efficiency.However, it depends on the correlation of the input data. A method of adding and outputting is also considered.

In the pre-prediction, the signal with the best encoding efficiency is a signal that is monotonically increased or monotonically decreased. This is because, at this time, the dynamic range of the prediction error becomes the smallest. Therefore, a method is conceivable in which a plurality of ways of extracting input data from blocks are prepared in different orders, and a method having the smallest prediction error dynamic range is selected. In this case, a code indicating a method of extracting input data (a kind of prediction method) is transmitted as one of the additional codes.

Although the above prediction method is extrapolation prediction, interpolation prediction may be used. For example, in the first embodiment, the input data is transmitted as it is without encoding only the head data of each block. Therefore, a value obtained by linear interpolation using the head data of each block is used as a prediction value, Is used as the prediction error. Since the interpolation prediction does not use the data decoded sequentially as in the extrapolation prediction for decoding the next data,
No error propagation occurs.

If the high frequency component in the input signal is very large, the input signal at a specific position in the block (for example, the head data of the block) or the average value of all data in the block is calculated as Is effective. For example, when the input signal is a repetition of 20, 30, 20, 30,..., The maximum value SX and the minimum value SN of the prediction error in the pre-prediction are 10, -10, respectively, and the dynamic range SDR is 20. On the other hand, if the head data 20 is a predicted value for all inputs in the same block, the maximum value SX and the minimum value SN of the prediction error are 10, 0, respectively, and the dynamic range SDR can be reduced to 10.

Effect of the Invention As described above, according to the present invention, it is possible to provide an encoding method and a decoding method capable of performing efficient transmission without anomalous errors with a simple configuration, and its practical value is large. .

[Brief description of the drawings]

FIG. 1 is a block diagram showing the configuration of a high-efficiency coding apparatus according to a first embodiment of the present invention, and FIGS. 2 (a) to 2 (c) are graphs showing examples of data in blocks having a one-dimensional configuration. FIG. 3 is a block diagram showing a configuration of a decoding apparatus according to the first embodiment, and FIG. 4 is a block diagram showing a configuration of a high-efficiency coding apparatus according to the second embodiment of the present invention and the configuration of the decoding apparatus. FIG. 5 is a block diagram showing a configuration of a high-efficiency encoding device according to a third embodiment of the present invention. FIG. 6 is a block diagram showing a configuration of a decoding device according to the third embodiment. FIG. 7 is a graph (waveform example) for explaining a method of reducing the dynamic range of the prediction error of the input signal by nonlinear processing, and FIGS. 8 (a) to (h) are the order of extracting data from the block and the prediction method. FIG. 9 shows a conventional encoding method M FIG. 10 (a) to (d) are operation waveforms (in a normal state) in the conventional MPCM system, and FIG. 11 is a quantization characteristic in the decoding device of the conventional MPCM system. Fig. 12 (a)-
(D) is a graph showing another example of an operation waveform (when an irregular error occurs) in the conventional MPCM system. 101 input terminal, 102 block divider, 103 delay unit, 104 register, 105 predictor, 106 subtractor, 107 minimum / maximum detection circuit, 108 subtractor, 109…
... Divisor data generator, 110 ... Remainder arithmetic unit, 111 ... Remainder encoder, 112 ... Output terminal of encoded data Ci, 113 ...
… Output terminal of divisor data OU, 114 …… Maximum value S of prediction error
X output terminal, 115 output terminal for input data D1, 201
... input terminal for encoded data Ci, 202 ... input terminal for divisor data OU, 203 ... input terminal for maximum value SX of prediction error, 204
…… Input terminal for input data D1, 205… Remainder decoder, 2
06 Adder, 207 Subtractor, 208 Offset regenerator, 209 Offset adder, 210 Switch, 211
... predictor 212 ... block decomposer 213 ... decoded data output terminal.

────────────────────────────────────────────────── ─── Continuation of front page (56) References IEICE Transactions, Vol. J 71-A, No. 2, P. 453-462 Proceedings of the 1988 IEICE Spring National Conference [Section A-1], page 1-22 Proceedings of the 1988 IEICE Spring National Conference (Section A-1) No. 1 -23 pages IEEE INTERNALON CONFERENCE ON COMMUNICATIONS '88 Vol. 2/3, p. 652-656 (58) Field surveyed (Int. Cl. ⁶ , DB name) H03M 7/36 H04N 7/00

Claims (27)

(57) [Claims] A block composed of a plurality of input data is used as a unit of encoding, and a maximum value SX and a minimum value SN of a prediction error, which is a difference between the input data in the block and a predicted value thereof, are obtained. −SN), a divisor data OU is used as a predetermined value, and a remainder E obtained by dividing each input data of the block by the divisor data OU is encoded, and the SX, SN,
An encoding method comprising transmitting an additional code including prediction error information regarding an OU and the encoded E together. 2. The apparatus according to claim 1, wherein the predicted value is input data in the same block as the data to be predicted.
Coding method as described. 3. The encoding method according to claim 1, wherein the predicted value is input data at a specific position in the same block as the data to be predicted. 4. The encoding method according to claim 1, wherein the predicted value is an average value of input data in the same block as the data to be predicted. 5. The encoding method according to claim 1, wherein interpolation prediction is used. 6. The encoding method according to claim 1, wherein prediction information indicating a prediction method having the highest encoding efficiency is used as one of the additional codes. 7. The method according to claim 1, wherein the divisor data OU transmitted as one of the prediction error information satisfies the following equation: OU = 2 ^M > SX−SN ≧ 2 ^M−1, where M satisfies an integer. Coding method as described. 8. The encoding method according to claim 1, wherein the input data at a specific position in the block is one of the additional codes. 9. The encoding method according to claim 1, wherein the average value of the input data in the block is one of the additional codes. 10. A signal obtained by converting an original input signal such that a dynamic range of a prediction error of the original input signal is equal to or less than a predetermined amount for each of a predetermined number of blocks.
Coding method as described. 11. The encoding method according to claim 10, wherein conversion information indicating a conversion method of the original input signal is transmitted as one of the additional codes. 12. The encoding method according to claim 10, wherein the conversion method of the original input signal multiplies the original input signal by a predetermined coefficient. 13. The encoding method according to claim 10, wherein the conversion method of the original input signal is to increase or decrease a part of the original input signal by a small amount. 14. The encoding method according to claim 10, wherein the conversion method of the original input signal performs high-frequency suppression on the original input signal. 15. A block composed of a plurality of input data is used as a unit of encoding, and a maximum value SX and a minimum value SN of a prediction error, which is a difference between the input data in the block and its predicted value, are obtained. −SN) to a given value greater than the divisor data OU
, The remainder E obtained by dividing each input data of the block by the divisor data OU is encoded, and the SX, S
N, a signal encoded by an encoding method of transmitting both the additional code including the prediction error information relating to the OU and the encoded E is input, and the divisor data OU is obtained from the transmitted additional code, and transmitted. The decoded remainder E is added to the divisor data OU of an integer N times as decoded decoded data, and the prediction error of the difference between the decoded data and its predicted value is obtained from the transmitted additional code. A decoding method, wherein the integer N is determined so as to satisfy a range. 16. The method according to claim 1, wherein the predicted value is input data in the same block as the data to be predicted.
5) The decoding method described. 17. The decoding method according to claim 15, wherein the predicted value is input data at a specific position in the same block as the data to be predicted. 18. The decoding method according to claim 15, wherein the predicted value is an average value of the input data in the same block as the data to be predicted. 19. The decoding method according to claim 15, wherein interpolation prediction is used. 20. The decoding method according to claim 15, wherein prediction information indicating a prediction method having the highest encoding efficiency is used as one of the additional codes. 21. The method according to claim 15, wherein the divisor data OU transmitted as one of the prediction error information satisfies the following equation: OU = 2 ^M > SX−2N ≧ 2 ^M−1 where M satisfies an integer. The decoding method as described. 22. The decoding method according to claim 15, wherein the input data at a specific position in the block is one of the additional codes. 23. The decoding method according to claim 15, wherein the average value of the input data in the block is one of the additional codes. 24. The decoding method according to claim 15, wherein the original input signal is reproduced by performing a reverse conversion of the decoded data on the basis of the conversion information in the additional code. 25. Decoding of coded data after occurrence of an error in coded data is performed by using decoded data in an adjacent block as a starting point of prediction and decoding in a direction different from that in the encoding. The decoding method according to claim 15, wherein 26. The decoding method according to claim 15, wherein the additional code having an error is decoded by using an additional code of an adjacent block. 27. If there is a region in which a prediction error obtained by performing a prediction across a block boundary using decoded data is out of the range of a prediction error obtained from an additional code, it is determined that a transmission error has occurred, and an adjacent block is determined. 16. The decoding method according to claim 15, wherein the decoding is performed by predicting the region in a direction different from the direction at the time of encoding, using the decoded data in the region as a starting point of prediction.