JP2002208860A - Device and method for compressing data, computer- readable recording medium with program for data compression recorded thereon, and device and method for expanding data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To completely restore original signal even by going through the processes of data compression and expansion and moreover to obtain satisfactory data compression rate and expanding device for sound signals, image signals, etc. SOLUTION: A band filter bank 3 generates a signal being the same as a signal of a full frequency band obtained, when a data expanding device side performs band synthesis, a subtracter 8 calculates an error signal between the signal of the total frequency band and the original signal inputted to a data compression device 1, and the compressed data of the error signal are outputted together with the compressed data of the signal of each frequency band. Thus, the data expanding device restores the signal of the total frequency band and the error signal on the basis of the compressed data, subtracts the error signal from the signal of the full frequency band, and completely restores the original signal inputted to the data compressing device 1.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data compression device and a data decompression device for audio signal or image signal data, and more particularly to a technique for completely restoring data before compression.

[0002]

2. Description of the Related Art In a conventional data compression method such as MPEG used in a compression apparatus for audio signal and image signal data, a sub-band signal of each band obtained by frequency-dividing an original signal is used. By performing the quantization process, the appearance probabilities of the sub-band signals in each band are biased, and then, for the sub-band signals in each band having a bias in the appearance probabilities, the bias of the data appearance probabilities is biased. A large amount of data has been compressed by performing variable-length coding such as Huffman coding using the technique described in Japanese Patent Application Laid-Open No. 5-31596.
No. 9, etc.).

[0003] On the other hand, the data compression method as described above has not been adopted in a field where high-quality audio and video signals need to be stored and distributed. This is because the normal data compression method as described above cannot completely restore the original audio signal or image signal data. That is, in the above-described method, a part of the information of the sub-band signal of each band is lost during the quantization process for rounding the value of the sub-band signal of each band. Also, an operation error occurs between the band division filter bank used for frequency division of the original signal and the band synthesis filter bank used for synthesizing the signals of the entire original frequency band. For this reason, even if each sub-band signal after the quantization process is combined with signals of all frequency bands by a band combining filter bank, the original audio signal and image signal cannot be completely restored. Therefore, there is a data compression / decompression device which can completely restore the original signal even after the compression / decompression process in a field where high-quality audio / video signals need to be stored and distributed. An image signal is obtained by obtaining a predicted value of a target pixel from a pixel value and performing a variable length encoding process such as Huffman encoding on an error (prediction error) between the predicted value and an actual pixel value. There is one that adopts the JPEG standard of the so-called reversible encoding system that compresses the data amount of the data.

[0004]

However, a data compression / decompression device employing the above-mentioned conventional lossless encoding method of the JPEG standard cannot obtain a good compression ratio, and the data is compressed and decompressed. In view of the necessity of adding a dedicated device to the system, there is a problem that sufficient cost-effectiveness cannot be obtained.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and it is possible to completely restore the original signal even after data compression / decompression processes, and to achieve good data compression. It is an object of the present invention to provide a data compression device and a data decompression device capable of obtaining a rate.

[0006]

According to a first aspect of the present invention, there is provided a band splitting means for orthogonally transforming an input digital signal and dividing the digital signal into a plurality of frequency band signals. Quantizing means for performing a quantization process for rounding a value of a predetermined digit or less for a signal value of each frequency band divided by the means, and according to an appearance probability of a signal of each frequency band after the quantization process by the quantization means. A first variable-length encoding unit that compresses the data amount of the signals in all frequency bands by encoding the signals in each frequency band while changing the code length of each signal. A compression unit that combines signals of the respective frequency bands after the quantization processing by the quantization unit to generate a signal of the entire frequency band, and a total frequency band generated by the band combination unit. Error signal calculating means for calculating an error signal which is a signal of a difference between the signal and the original digital signal input to the band dividing means, and according to the appearance probability of the value of the error signal calculated by the error signal calculating means. A second variable-length encoding unit that encodes the error signal while changing the code length of each signal value, thereby compressing the data amount of the error signal. The data of the signal of each frequency band compressed by the converting means and the data of the error signal compressed by the second variable length coding means are output.

In the above configuration, the input digital signal is orthogonally transformed by the band dividing means to divide the signal into a plurality of frequency bands, and to concentrate the generation of the signal in a specific frequency band. The quantization means rounds the signal value of each frequency band to a value less than a predetermined digit. The first variable length encoding means compresses data using the appearance probability of the signal of each frequency band to the signal of each frequency band after the quantization processing in which the appearance probability of the signal is biased by this frequency band. By doing so, a good data compression ratio can be obtained.

Next, the signals of the respective frequency bands quantized by the above-described quantization means are synthesized into signals of all the frequency bands by the band synthesis means, and the signals of the entire frequency band and the elements inputted to the band division means are combined. An error signal which is a signal of a difference from the digital signal is calculated by the error signal calculating means, and the data of the error signal is subjected to data amount compression processing by the second variable length coding means. Then, it outputs the data of the signal of each frequency band compressed by the first variable length coding means and the data of the error signal compressed by the second variable length coding means.

The above data compression device is used in combination with the data decompression device according to claim 8. The band synthesizing means of this data decompression device has the same configuration as the band synthesizing means of the data compression device. When the data decompression device receives the data of the signal of each frequency band and the data of the error signal, which have been encoded and compressed by the data compression device, the data decompression device converts the encoded signal of each frequency band into the first signal. And the encoded error signal is decoded into the original error signal by the second decoding means. Then, the decoded signals of the respective frequency bands are combined into signals of all the frequency bands by the band combining means. However, the signal of each frequency band serving as a synthesis source includes an error due to the rounding process of the signal value of each frequency band by the quantizing means on the data compression device side, and the signal of the signal by the band dividing means on the data compression device side. Since the calculation error generated at the time of the orthogonal transformation is included, the error is also included in the signals of all the frequency bands obtained by combining the signals of these frequency bands. In addition, the signals in the entire frequency band include an arithmetic error generated in the signal combining process by the band combining means. The data decompression device subtracts the original error signal decoded by the second decoding unit from the signals in the entire frequency band including these errors using the error removing unit, and outputs a signal in the entire frequency band without errors. . Thus, even after the data compression by the data compression device and the data decompression by the data decompression device, the same signal as the original digital signal input to the data compression device can be obtained.

[0010] Further, a prediction means for predicting a present digital signal from a past digital signal, and a prediction error signal for calculating a prediction error signal which is a signal of a difference between the digital signal predicted by the prediction means and an actual digital signal. A calculation unit may be further provided, and the prediction error signal calculated by the prediction error signal calculation unit may be input to the band division unit and the error signal calculation unit. Thereby, when the digital signal to be subjected to data compression is an easily predictable signal such as a music signal, a better data compression ratio can be obtained as compared with a case where the data of the digital signal is directly compressed. .

The digital signal input to the band dividing means and the error signal calculating means may be an audio signal. When an audio signal is orthogonally transformed by band dividing means and divided into signals in a plurality of frequency bands, signals are generated concentrated in a specific frequency band. The data compression of the audio signal can be increased by compressing the data of the subsequent audio signal by using the bias of the appearance probability of the signal of each frequency band by the first variable length encoding means. it can.

The digital signals input to the band dividing means and the error signal calculating means are signals of luminance, color difference and the like in the image divided into small areas, and the band dividing means performs the processing based on these signals. A signal of a spatial frequency may be generated, and the signal of the spatial frequency may be divided into signals of a plurality of frequency bands. Signals such as luminance and color difference in an image (hereinafter referred to as image signals) often change gently. If a signal having such a gradual change is divided into signals in a plurality of spatial frequency bands by band dividing means, low signals are generated. The signal concentrates on the frequency band. For the data of the image signal after the band division in which the appearance probability of the signal is biased by the frequency band, the first variable length encoding means compresses the data using the bias of the appearance probability of the signal in each frequency band. By doing so, the data compression rate of the image signal can be increased.

According to a fifth aspect of the present invention, there is provided a method for compressing the data amount of an input digital signal, comprising the steps of orthogonally transforming the input digital signal to divide it into signals of a plurality of frequency bands. Performing a quantization process of rounding a value of a predetermined digit or less for a signal value of each divided frequency band, and a code of each signal according to an appearance probability of a signal of each frequency band after the quantization process. A step of compressing the data amount of the signals of all frequency bands by encoding the signals of each frequency band while changing the length, and synthesizing the signals of each frequency band after the quantization process to obtain Generating a signal of a band, and calculating an error signal that is a signal of a difference between the generated signal of the entire frequency band and a digital signal that is a source of orthogonal transform. Compressing the data amount of the error signal by encoding the error signal while changing the code length of each signal value according to the appearance probability of the value of the error signal. Outputting the data of the band signal and the data of the compressed error signal. According to this method, the same operation as that of the first aspect can be obtained.

The method according to claim 5, further comprising the step of predicting a current digital signal from a past digital signal, and a prediction error signal being a signal representing a difference between the predicted digital signal and an actual digital signal. , And the prediction error signal calculated in the calculating step may be used as a digital signal that is a conversion source of the orthogonal transform in the dividing step. Thereby, the same operation as in the second aspect can be obtained.

According to a seventh aspect of the present invention, there is provided a computer-readable recording medium having recorded thereon a program for compressing a data amount of a digital signal by a computer, wherein the program is stored in a storage medium. A program for data compression in which any of the data compression methods described above is executed. In this configuration, the same operation as described above can be obtained by causing the computer to read the data compression program and executing the program on the computer.

According to an eighth aspect of the present invention, there is provided a data decompression device for decompressing data compressed by the data compression device according to the first aspect, wherein the encoded signal of each frequency band is converted into an original signal. First decoding means for decoding into signals in each frequency band, band synthesis means for synthesizing signals in each frequency band after decoding by the first decoding means to generate signals in all frequency bands, Second decoding means for decoding the error signal into the original error signal, and error removing means for subtracting the error signal decoded by the second decoding means from the signal of the entire frequency band generated by the band synthesizing means. It is provided.

In this configuration, the data of the signal of each frequency band encoded by the data compression device according to the first aspect is decoded by the first decoding means into the data of the signal of the original frequency band. By combining these signals of each frequency band by the band combining means, signals of all frequency bands are generated. Further, the error signal encoded by the data compression device according to the first aspect is decoded into the original error signal by the second decoding means. Then, the error signal decoded by the second decoding unit is subtracted from the signal of the entire frequency band synthesized by the band synthesizing unit. This makes it possible to remove the error from the signals of all the frequency bands obtained by synthesizing the signals of the respective frequency bands including the error, so that the same signal as the original digital signal input to the data compression device is obtained. be able to.

According to a ninth aspect of the present invention, there is provided a data decompression method for decompressing data compressed by the data compression method according to the fifth aspect of the present invention. Decoding into frequency band signals, combining the decoded frequency band signals to generate all frequency band signals, and decoding the encoded error signal into the original error signal. , Subtracting the decoded error signal from the signal of the entire frequency band generated by the generating step. According to this method, the same operation as that of the eighth aspect can be obtained.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a data compression device and a data decompression device according to one embodiment of the present invention will be described with reference to the drawings. With reference to FIG. 1, a data compression apparatus and a data compression processing method according to the present embodiment will be described by taking a case where audio data is compressed as an example. The data compression device 1 is similar to a data compression device based on a conventional method such as MPEG.
A band dividing filter bank 3 (band dividing means) for dividing digital signal data (hereinafter abbreviated as PCM data) input from PCM (Pulse Code Modulation) data input means 2 into a plurality of frequency band signals; A quantizing device 4 (quantizing means) for rounding a value of a predetermined digit or less of a signal (hereinafter referred to as a sub-band signal) band-divided by the filter bank 3, and each sub-band output from each quantizing device 4 A sub-band signal encoding device 5 (first variable-length encoding means) for performing variable-length encoding such as Huffman encoding on a signal is provided. The data compression device 1 includes, in addition to the above devices, a band synthesis filter bank 7 (band synthesis means) that is not included in the conventional data compression device.
A subtractor (error calculating device: error signal calculating means in claims) 8 and an error signal coding device 9 (second variable length coding means) are provided. The band synthesis filter bank 7
This is a device that combines the sub-band signals after digit rounding processing by the quantization device 4 to generate signals in all frequency bands. The subtracter 8 calculates an error between the signal of the entire frequency band generated by the band synthesis filter bank 7 and the original PCM data signal input from the PCM data input means 2. Error signal encoding device 9
Is for performing variable length coding processing such as Huffman coding on the error signal output from the subtractor 8 to compress the data amount of the error signal.

The PCM data input means 2 for supplying PCM data to the data compression device 1 is a device comprising a microphone, an amplifier, an A / D converter and the like when compressing audio data in real time. In the case of compressing audio data that has been recorded once, the apparatus serves as a recording signal reproducing apparatus. However, if the recording signal reproducing device is of an analog type, an A / D converter is further required.

Next, data compression processing by the data compression apparatus 1 will be described. The digital audio signal data supplied by the PCM data input means 2 is sent to a band division filter bank 3 in the data compression device 1, and the band division filter bank 3 generates a plurality of narrow band signals (sub-band signals). Is divided into At this time, the band division filter bank 3 divides the line spectrum included in the audio signal into as many bands as can be sufficiently separated. As a method of band division, a method of dividing the entire band at equal intervals may be used,
Further, a method of dividing the entire band at irregular intervals may be used. Here, the entire band is divided into N equally-spaced bands. Assuming that the sampling frequency of the audio signal is f _s , the total bandwidth W of the audio signal is represented by the following equation (1). If _{W = f s / 2 ··· (} 1) Why, on the basis of sampling information of the certain signal, in order to reproduce the waveform of the original signal is sampled at a frequency higher than twice the highest frequency of the original signal This is because there is a theorem that must be done. This theorem is called the sampling theorem.

When the audio signal having the bandwidth W is divided into bands by the band division filter bank 3, the bandwidth of each subband signal becomes W / N. That is, the bandwidth of each sub-band signal after band division is 1 / N of the total bandwidth W of the original audio signal. Here, the above (1)
According to the equation, since a proportional relationship is established between the bandwidth W of the signal to be sampled and the frequency f _{s of the} sampling signal, the bandwidth of each subband signal after band division and the original sound The ratio to the bandwidth of the signal is 1 /
In consideration of N, even if the sampling frequency f _s ′ for each subband signal after band division is set to 1 / N of the sampling frequency f _s of the original audio signal, the information of the original audio signal is not lost. Absent. Therefore, the sampling frequency f _s ′ for each sub-band signal required to restore the original sub-band signal is expressed by the following equation (2). f _s' = from f _s / N ··· (2) the above equation, the number of samples f _T of the signal of the entire band per second (N number of bands) is represented by the following formula. _{f T = Nf s' = f} s ··· (3) above (1) and from (3), the number of samples of the signal per second before and after band division is no change as a whole I understand.

Here, the above-mentioned band division filter bank 3
The reason why band division is performed by using is described. Considering the characteristics of audio signals, normal audio signals, especially music signals, have a continuous waveform when the signal within a certain period of time is viewed with the time axis as the horizontal axis, but when viewed with the frequency as the horizontal axis,
It is often a collection of several line spectra. For this reason, when the audio signal is divided into signals in a narrow frequency band (hereinafter, referred to as a subband) by the band division filter bank 3, the signal is concentrated on a specific subband. Therefore, the apparent entropy of the audio signal after the band division (for the information comprising a plurality of messages, the average per one message when the number of bits is reduced to the utmost with care taken so as not to impair the content of the information at all) It can be expected that the information amount (average number of bits)) is lower than the entropy of the audio signal on the horizontal axis on the time axis before band division. Therefore, a compression method (entropy coding method (variable length coding method) using the bias of the appearance frequency of a symbol (a value of one sample of a signal) such as Huffman coding or arithmetic compression is applied to the audio signal data after band division. ), The compression ratio is generally higher than when the audio signal before band division is compressed by the same compression method. Therefore, the audio signal is subjected to band division processing before being encoded and compressed.

When the above band division processing is completed, the band division filter bank 3 outputs each subband signal to the quantization device 4. Upon receiving each of the sub-band signals, the quantization device 4 performs a process of rounding a number below a certain digit in these signal values to reduce the information amount of each of the sub-band signals. In general quantization processing in irreversible compression, a quantization device often changes a digit to be rounded for each band or a digit to be rounded temporally in accordance with human auditory characteristics. . However, the present invention
Since the purpose is to prevent physical deterioration of the audio signal, not to prevent the sound quality from deteriorating in audibility, the quantizing device 4 fixes the digits to be rounded.

When the digit rounding process by the quantization device 4 is completed, the entire band audio signal and the PCM data input means synthesized on the basis of each sub-band signal after the digit rounding process output from the quantization device 4. A signal of an error between the original audio signal and the original audio signal input from step 2 is generated. The reason for generating such an error signal is that each of the sub-band signals after the digit rounding process output from the quantization device 4 is synthesized by a band synthesis filter bank (see FIG. 2) on the data decompression device side. This is because, when the audio signal in the frequency band is restored, the restored signal has a slight error as compared with the original audio signal. This error is caused by an error (quantization error) caused by the digit rounding process performed by the quantization device 4 and an operation error in the band division filter bank or the band synthesis filter bank. Therefore, in order to remove such an error and completely restore the original audio signal, the data compression device 1
However, in addition to the compressed data of each sub-band signal containing an error, a signal (error signal) for removing the error is also output.

The process of generating the error signal is specifically performed as follows. The band synthesis filter bank 7 receives each of the sub-band signals after the digit rounding process output from the quantization device 4 and synthesizes these signals. The subtractor (error calculation device) 8 receives the signal obtained by synthesizing all the sub-band signals from the band synthesis filter bank 7, receives the original audio signal from the PCM data input means 2, and converts the signal obtained by synthesizing all the sub-band signals. An error signal is calculated by subtracting the original audio signal.

Each of the sub-band signals quantized by the above-described quantization device 4 is encoded by a sub-band signal encoding device 5, and the error signal obtained by the subtracter 8 is encoded by an error signal encoding device 9. Encoded. The encoding method used for these encoding processes is an encoding method that utilizes a bias in the appearance frequency of symbols, such as Huffman encoding or arithmetic compression encoding. The encoded subband signal data and error signal data are output to the compressed data output means 6 together. As for the error signal, since the power of the signal before encoding is small (the signal is weak), the data amount after encoding is small and sufficiently smaller than the data amount of the subband signal after encoding. Therefore, even if the encoded error signal data is output together with the encoded subband signal data, the data amount of the entire output data hardly increases.
The compressed data output means 6 becomes a storage device when storing audio signals, and becomes a communication line when distributing audio signals.

Next, as described above, the audio signal is divided into bands.
Compression after compressing the audio signal directly
FIGS. 3A to 3E show that a high compression ratio can be obtained.
This will be described with reference to FIGS. FIG. 3 (a)
Input to the 4-output band splitting filter bank in FIG.
FIG. 3 shows a signal waveform (hereinafter abbreviated as an input waveform), and FIG.
(B), (c), (d), and (e) are the same as those in FIG.
Each sub-bar output from the 4-output band division filter bank
Command signal S₀, S₁, S₂, S₃3 shows the waveforms of FIG. Also,
FIG. 4A shows one of the band division filter banks 3 in FIG.
FIG. 4 shows a configuration of an example 4-output band division filter bank;
FIG. 4B shows each filter of the four-output band division filter bank.
Frequency characteristics of the filter and the two frequencies included in the input waveform
f₁, F_Two(F ₁<F_Two2) Line specs corresponding to the components
Torr. 3 (a), 3 (b) and 3 (d),
Each vertical line represents the waveforms 11, 14, 17 in the respective figures.
Are shown as sampling values 12, 15, and 18, respectively. This
Here, as a simple example of an audio signal, two frequencies f₁When
f_Two(F₁<F_TwoConsider the audio signal containing only the component
Then, the time waveform of this audio signal is as shown in FIG.
(In FIG. 3A, f _Two= 5f₁Here is an example of
are doing). Also, the four output bands shown in FIG.
The divided filter bank 3a has F₀~ F₃Four fills
Data. As shown in FIG.
Filter F₀~ F₃Is lower for lower numbered filters
Pass signals of different frequencies.

Here, a waveform 11 shown in FIG.
The audio signal is divided into four output band division signals as shown in FIG.
It is assumed that the input is made to the filter bank 3a. FIG. 4 (b)
As shown in the figure, two frequencies included in the input waveform 11
Number f₁And f_TwoAnd the line spectra L corresponding to₁And L
₂And the filter F₀And filter F_TwoAnd pass
Filter F₀~ F₃Each divided by
Subband signal S₀~ S₃The output waveforms of FIG.
Waveforms 14, 16, 17, 19 shown in (b) to (e)
become that way. Audio signal has 4 output band division filter band
Downsampling to 1/4 when passing through 3a
3 (b) and 3 (d).
The numbers of the pulling values 15 and 18 both correspond to the input waveform 11.
サ ン プ リ ン グ of the number of sampling values 12 obtained. Paraphrase
And each output waveform 1 shown in FIGS. 3B and 3D.
The sampling frequency for 4,17 is the same as the original input
This is １ ／ of the sampling frequency for the waveform 11.
Frequency f included in input waveform 11₁The component of is S₀only
And the frequency is f_TwoThe component of is S_TwoOnly appearing in S₁When
S _ThreeDoes not show a signal. Thus, time continuous
Even if a waveform is input, the signal is localized when divided into bands.
I do.

Therefore, the sub-band signal S ₀ after the band division
Who to S ₃ are entropy apparent than the signal of the entire frequency band indicated by the original input waveform 11 is low. Therefore, when performing compression by applying Huffman coding or arithmetic compression coding, it is better to perform coding after dividing the band than to directly code the input signal of the entire frequency band. Obtainable. Signal represented by the input waveform 11 is with only two line spectra of L ₁ and L _2, the actual speech signal contains more line spectrum. Therefore, in order to separate those line spectra, it is necessary to divide them into narrower frequency bands.

Next, the operation of the data decompression device for decompressing the data compressed by the data compression device 1 will be described with reference to FIG. 2 by taking as an example a case where the decompression device is applied to the compression of audio data. I do. Here, decompression means that the compressed data is restored to the original PCM data.
The compressed data input means 22 is for inputting the data of the audio signal compressed by the data compression device 1 to the data decompression device 21. This compressed data input means 22
Is a storage device when inputting compressed data stored in a recording device, and is a communication line when inputting compressed data distributed using a communication line. The compressed data input from the compressed data input unit 22 is separated into a sub-band signal and an error signal, and the sub-band signal decoding unit 23 (first decoding unit) and the error signal decoding unit 25 (second decoding unit). Means). The subband signal decoding device 23 decodes the encoded subband signal of each band input from the compressed data input means 22 into the original subband signal, and
4 (band combining means). The band synthesis filter bank 24 synthesizes the sub-band signals of each of these bands to generate a single full-band audio signal.

On the other hand, the error signal decoding device 25 decodes the encoded error signal input from the compressed data input means 22. The audio signal of all bands generated by the band synthesis filter bank 24 and the error signal restored by the error signal decoding device 25 are output to a subtracter 26 (error removing means). The subtracter 26 includes a band synthesis filter bank 24
Signal decoding apparatus 2 from the audio signal of all bands synthesized by
By subtracting the error signal restored in step 5, the error is removed from the audio signal of the entire band obtained by the synthesis, and the original audio signal is completely reproduced. The reproduced audio signal is
It is output to the M data output means 27. The PCM data output unit 27 is a device including a D / A converter, an amplifier, a speaker, and the like when a reproduced sound is directly heard by a person, and when the reproduced sound is temporarily stored,
It becomes a storage device. In order to completely expand the data of the compressed audio signal to the original audio signal, the band synthesis filter bank 7 of the data compression device 1 and the band synthesis filter bank 24 of the data expansion device 21 need to generate an operation error. It is necessary to have exactly the same characteristics including the way of doing.

Next, a description will be given of a process in the case where the data compression device 1 and the data decompression device 21 are applied to compression and decompression of image signal data. When a signal of a black and white image is generated, as shown in FIG. 5, grid-like coordinates are taken on a black and white image 31, and the luminance value I (x, y) is calculated at each (x, y) coordinate 32. In many cases, sampling is performed and the sampled value is used as PCM data. In the example shown in the figure, the monochrome image 31 is divided into X pixels in the horizontal direction and Y pixels in the vertical direction. When the data of the image signal of the monochrome image 31 is compressed by the data compression device 1,
As shown in FIG. 6, the PCM data input means 2 shown in FIG. 1 divides the black-and-white image 31 into rectangular small areas 33 each having M pixels in the horizontal direction and N pixels in the vertical direction. Are sequentially input to the band division filter bank 3 shown in FIG. Band division filter bank 3
Divides the signal of the luminance value of the given small region 33 into subband signals based on the spatial frequency. The spatial frequency in this case is a value representing the degree of change in luminance between adjacent pixels in terms of frequency. The PCM data input means 2 for compressing the data of the image signal, when compressing the read image signal in real time,
It is composed of a scanner or the like, and when data of an image signal once recorded is compressed, it becomes a playback device of a recorded signal.

FIGS. 7A and 7B show how the signal of the luminance value in the small area 33 is divided by the band division filter bank 3 into subband signals based on the spatial frequency. As shown in FIG. 7A, the band division filter bank 3
A signal of the luminance value of each pixel 34 in the small area 33 having M pixels in the horizontal direction and N pixels in the vertical direction is input. The band division filter bank 3 converts these luminance value signals into signals in the spatial frequency domain 35, and as shown in FIG.
Signal 37 of subband 36 of M in the horizontal direction and N in the vertical direction
Divided into In this case, since the small area 33 is two-dimensional, the spatial frequency area 35 is also two-dimensional. The spatial frequencies of the sub-bands 36 in the spatial frequency domain 35 are lower in frequency in the horizontal direction, higher in frequency toward the right side, and higher in frequency in the vertical direction. The lower the frequency, the higher the lower the frequency. As shown in FIG. 8, the PCM data input means 2 shown in FIG.
3 is scanned, and the signal of the luminance value of each pixel 34 in each small area 33 is sequentially sent to the band division filter bank 3. When the band division filter bank 3 Output a signal corresponding to. Generally, when an image signal such as a luminance value is converted into a signal in the spatial frequency domain 35 and observed, it is known that signal power is concentrated in a low frequency band. Therefore, in the case of FIG. 7B, a strong signal concentrates on the upper left subband 36. in this way,
When the image signal is converted into a signal in the spatial frequency domain 35 and divided into narrow band sub-band 36 signals, the bias of the signal strength becomes larger than that of the original image signal. By performing compression utilizing information bias such as Huffman coding or arithmetic compression coding, a good compression ratio can be obtained. Also in the case of compressing image signal data, the sub-band signals output from the band division filter bank 3 are synthesized by the band synthesis filter bank 7 and compared with the original image signal (by subtracting the original image signal). The mechanism for obtaining the error signal is the same as that for compressing the audio signal data. The mechanism for expanding the data of the image signal is the same as the mechanism for expanding the data of the audio signal, except that the band synthesis filter bank 24 is a band synthesis filter of a two-dimensional spatial frequency. In the case of a color image, the image signal is decomposed into three color components such as R, G, B, Y, Cb, and Cr, and the same data as in the case of the above-described monochrome image 31 is obtained for each color component. What is necessary is just to perform a compression process.

In the above description, for simplicity,
Although the example in which the PCM data is directly divided into bands and compressed is shown, a prediction filter predicts the current PCM data from the past PCM data, and obtains the predicted value and the actual PCM data.
An error from the M data (hereinafter referred to as a prediction error) may be compressed by the data compression device 1 shown in FIG. Especially for audio compression, if the prediction filter is properly designed,
A better compression ratio can be obtained by compressing the data of the prediction error than by directly compressing the CM data.

Next, an embodiment in which a prediction filter is used in a stage preceding the audio signal data compression device 1 and a stage following the data decompression device 21 will be described with reference to FIGS. FIG. 9 shows a prediction error calculator 40 including a prediction filter provided at a stage prior to the data compression device 1 shown in FIG.
Is shown. FIG. 10 shows a configuration of a PCM data restoration unit 45 including a prediction filter provided at a subsequent stage of the data decompression device 21 shown in FIG. The prediction error calculation unit 40
A delay element 42 for holding past PCM data output from the PCM data input means 2; a prediction filter 41 (prediction means) for predicting current PCM data based on past PCM data output from the delay element 42; From the predicted value of the current PCM data output from the prediction filter 41, the current actual P output from the PCM data input means 2.
Subtractor 4 for calculating the prediction error by subtracting the CM data
3 (prediction error signal calculation means). Also, P
The CM data restoration unit 45 includes a delay element 47 that holds the past PCM data output from the subtracter 26 in FIG.
Prediction filter 48 for predicting current PCM data based on past PCM data output from delay element 47
And the prediction error output from the subtracter 26 is subtracted from the predicted value of the PCM data output from the prediction filter 48,
And a subtractor 46 for calculating current actual PCM data. The prediction filter 41 and the prediction filter 48
Means that those having the same characteristics must be used.

9 and 10, the data of the audio signal input from the PCM data input means 2 is sequentially held in each delay element 42. The prediction filter 41 calculates a predicted value of the current PCM data based on the data of the past several times of the audio signal held in each of the delay elements 42,
This predicted value is output to the subtractor 43. The subtracter 43 calculates a prediction error by subtracting the current actual PCM data output from the PCM data input means 2 from the predicted value of the current PCM data output from the prediction filter 41, and calculates the prediction error in FIG. To the indicated data compression device 1. After the data of the prediction error is compressed by the data compression device 1, the data is sent to the data decompression device 21 shown in FIG. 2 via a storage device or a communication line, and is restored to the original prediction error data. The reconstructed prediction error data is supplied from the subtractor 26 to P
The delay data is sent to the CM data restoring unit 45 and sequentially
7 is held. The prediction filter 48 determines the current PCM based on the past PCM data output from the delay element 47.
The data is predicted, and the predicted value of the PCM data is output to the subtractor 46. The subtracter 46 subtracts the prediction error output from the subtractor 26 from the predicted value of the PCM data,
The original PCM data input from the PCM data input means 2 is restored.

Next, the prediction error calculating section 40 and the PCM data restoring section 45 shown in FIGS. 9 and 10 are replaced with the data compression apparatus 1 and the data decompression apparatus 21 shown in FIGS.
11 and FIG.
This will be described with reference to FIG. In this embodiment, a 256-point DCT (DiscreteCo
sine Transform: 256-point IDCT (Inverse Discrete)
A dynamic Huffman coding device is used as a Cosine Transform (inverse discrete cosine transform) device, a subband signal coding device 5 and an error signal coding device 9, and data compression is performed using an input one sample before as a current prediction value. Here, the prediction filter used for calculating the predicted value is one delay element 42
It is not shown in the figure because the input from the device is output as it is. First, the data compression device 1 will be described. The dynamic Huffman coding method used in the subband signal coding device 5 and the error signal coding device 9 is based on the statistics of the distribution of the appearance frequency of symbols while coding. take,
This is a Huffman coding method for updating a coding table according to statistics. In the normal Huffman coding (static Huffman coding) method, it is necessary to grasp the appearance frequency distribution of symbols in advance.

On the other hand, in the dynamic Huffman coding method, a suitable appearance frequency distribution is assumed at first, statistics are obtained while coding, and the coding table is updated according to the statistical values, thereby gradually increasing the actual frequency. Can be appropriately encoded according to the appearance frequency distribution of. The dynamic Huffman coding scheme is initially inefficient. However, when encoding many symbols, the initial inefficiency is negligible as a whole. Data compression device 1
When strictly performing static Huffman coding, it is necessary to read a signal (symbol sequence) not only when actually performing coding, but also when obtaining statistics of the symbol appearance frequency distribution. Therefore, the signal needs to be read twice in the static Huffman coding method, but only once in the dynamic Huffman coding method. In the case of the static Huffman coding method, the data compression device 1 needs to add an encoding table to compressed data and output the compressed data. In the case of the dynamic Huffman coding method, the data compression apparatus 1 No need to add.

In this embodiment, since a 256-point DCT is used for the band division filter bank 3, the sub-band signal has 256 channels. When the sampling frequency for the PCM signal is 44.1 kHz, the total bandwidth W of the audio signal is 22.05 kHz according to the above equation (1).
Therefore, in this case, the bandwidth per channel is about
86Hz. The 256 quantizers 4 round the value of each sub-band signal output from the DCT unit to three bits or less after the decimal point to convert it to a signal with two bits after the decimal point.
Although dynamic Huffman coding is used for encoding the subband signal and the error signal, the distribution of the symbol appearance frequency differs between the subband signal and the error signal. Use a different encoding table. Also, since the distribution of the symbol appearance frequency differs depending on the channel of the subband signal, a different coding table is used for each channel.

Next, the data decompression device 2 shown in FIG.
Explaining No. 1, the dynamic Huffman decoding device used as the subband signal decoding device 23 uses the same encoding table as the subband signal encoding device 5 (FIG. 11). Also, the dynamic Huffman decoding device used as the error signal decoding device 25 uses the same encoding table as the error signal encoding device 9 (FIG. 11). Further, the band synthesis filter bank 24 is a 256-point IDCT unit having completely the same characteristics including the calculation accuracy as the band synthesis filter bank 7 (FIG. 11).

The data compression device 1 and data decompression device 21 shown in FIGS. 11 and 12 are composed of a band division filter bank 3, a quantization device 4, a subband signal encoding device 5, a band synthesis filter bank 7, a subtraction Unit 8, error signal encoding unit 9, sub-band signal decoding unit 23, band synthesis filter bank 24, error signal decoding unit 25, subtractor 2
6, the subtractor 46, the delay element 47, etc. are implemented as individual devices.
(Digital Signal Processor) or the like may be implemented by software.

Next, the data compression device 1 (FIG. 11) and the data decompression device 21 (FIG. 12) described above with reference to FIG. 13 are realized by software on a personal computer, and music on a commercially available music CD is reproduced. The compression result will be described. A recording medium such as a CD-ROM in which the data compression software program is recorded in this case corresponds to a computer-readable recording medium in which the data compression program is recorded. The compression ratio of each music in the figure is obtained by the following equation (4). Compression rate = data capacity before compression / data capacity after compression × 100 (4) Further, the average compression rate in the figure is obtained by simply averaging the compression rates of all the music pieces in the figure. . The sampling frequency of PCM data was set to 44.1 kHz for all songs. The data compression device 1 employs a two-channel stereo system, and sets the number of quantization bits for each of the left and right channels to 16 bits. The compression and decompression are performed independently for the left channel input signal and the right channel input signal, and the correlation between the left and right channel signals is not used for compression. As shown in the figure, there is a difference in the data compression ratio depending on the music. However, on average, the data compression ratio is about 57%. When compression was attempted for dozens of songs in various genres, MEGALITH in the figure was the worst compression ratio, and NADIE was the best compression ratio.

FIG. 14 shows the intensity distribution (spectrum) of the subband signal of MEGALITH and FIG. 15 shows that of NADIE. In both figures, the left channel D at 58.0000 seconds from the beginning of the song
This shows the processing result of CT. A comparison between the two figures shows that in MEGALITH with a low compression ratio, a sub-band signal stronger than in NADIE with a high compression ratio exists over a wide band. As shown in this example, when the distribution of the signal distribution is large and the signal is weak, a better compression ratio can be obtained. After performing data compression and data decompression on the audio signals of all songs, the original PC
M data is restored.

Next, an embodiment of an image signal data compression apparatus will be described. FIG. 16 shows a block configuration of the data compression device 1 of the present embodiment. In this embodiment, an 8 × 8-point DCT unit as the band division filter bank 3, an 8 × 8-point IDCT unit as the band synthesis filter bank 7, a static Huffman code as the subband signal encoding device 5 and the error signal encoding device 9. And a prediction filter are not used. Here, the target image includes 320 (horizontal) × 240 (vertical)
A monochrome image of about pixels is assumed. The PCM data input means 2 divides the PCM data representing the luminance value of the image into small areas of 8 × 8 pixels. By this division processing, for example,
If the image to be compressed is composed of 320 pixels horizontally by 240 pixels vertically, the image is divided into 1200 small areas of 40 in the horizontal direction and 30 in the vertical direction. Next, the PCM data input means 2 scans the small area 33 zigzag by the method shown in FIG. 8 and sequentially sends the luminance value signals of the small area 33 to the band division filter bank 3 and the subtractor 8. The band division filter bank 3 performs an 8 × 8 point two-dimensional DCT based on the signal of the luminance value of each small region 33, and converts the signal of the luminance value of each small region 33 into a spatial frequency of 64 subbands. The signal is divided into signals, and these signals are sent to the quantization device 4. The quantization device 4 performs a quantization process on each of the transmitted subband signals, that is, a process of rounding the signal to three bits or less after the decimal point and approximating the signal to a numerical value to two bits after the decimal point. The subband signal quantized by the quantization device 4 is sent to the subband signal encoding device 5 and the band synthesis filter bank 7.

The subband signal encoding device 5 performs a static Huffman encoding process on the subband signal sent from the quantization device 4. On the other hand, the band synthesis filter bank 7
Performs an 8 × 8-point two-dimensional IDCT on the sub-band signal sent from the quantization device 4 and reproduces a spatial frequency signal divided into sub-bands into a signal of a luminance value of the entire band,
It is sent to the subtractor 8. The subtractor 8 subtracts the original luminance value signal sent from the PCM data input means 2 from the reproduced luminance value signal, calculates an error signal, and sends it to the error signal encoding device 9. The error signal encoding device 9 performs a static Huffman encoding process on the error signal sent from the subtractor 8.

Since the subband signal and the error signal have different symbol frequency distributions, different coding tables are used for Huffman coding of both. Further, in the case of a sub-band signal, the distribution of the appearance frequency of the symbol is different for each sub-band, so the sub-band signal encoding device 5 uses a different encoding table for each sub-band. When performing static Huffman coding, it is usually necessary to add a coding table to the compressed information and output the information. But,
In this embodiment, a total of 65 coding tables, one for error signal coding and 64 for subband signal coding, are used. The size of the table cannot be ignored. This problem can be avoided by assuming a standard symbol appearance frequency distribution in advance and determining a standard encoding table, and using this standard encoding table during data compression and data decompression. . The standard symbol appearance frequency distribution is obtained by processing a large number of images in advance, finding the symbol appearance frequency distributions for all the images, and averaging these appearance frequency distributions.

The reason for performing dynamic Huffman coding when compressing audio signal data and performing static Huffman coding when compressing image signal data is as follows. This is because the number of symbols to be coded is too small in the image of the degree, and the statistical value of the symbol appearance frequency distribution does not sufficiently converge when dynamic Huffman coding is performed. When compressing data of a larger image, the data may be compressed by a dynamic Huffman coding process.

FIG. 17 shows a block diagram of a data decompression device 21 corresponding to the above-described data compression device 1 (FIG. 16). The static Huffman decoding device used as the subband signal decoding device 23 uses the same encoding table as the subband signal encoding device 5 in FIG. The static Huffman decoding device used as the error signal decoding device 25 uses the same encoding table as the error signal encoding device. Furthermore, the band synthesis filter bank 24 is a two-dimensional 8 × 8 point having completely the same characteristics as the band synthesis filter bank 7 including the calculation accuracy.
IDCT device.

As described above, according to the data compression apparatus 1 and the data decompression apparatus 21 of the present embodiment, the same signal as the signal of the entire frequency band obtained when the data decompression apparatus 21 performs band synthesis. The data compression apparatus 1 generates a signal of an error between the signal and the original digital signal input to the data compression apparatus 1, and calculates the compressed data of the error signal together with the compressed data of the signal of each frequency band. Added output. As a result, the data decompression device 21
By subtracting the restored error signal from the restored signal in the entire frequency band, the original signal in the entire frequency band input to the data compression device 1 can be completely restored. Also, the input digital signal is subjected to discrete cosine transform (DCT) by the band division filter bank 3 to divide it into a plurality of sub-band signals, and then perform quantization processing on each of the divided sub-band signals. By doing so, it is possible to concentrate signal generation on a specific subband. A good data compression ratio can be obtained by performing Huffman encoding processing, which is data compression processing using the level of signal appearance probability, on data having a bias in signal appearance probability due to this subband. it can.

The present invention is not limited to the above embodiment, and various modifications are possible. For example, in each of the above embodiments, the case where the signal of the luminance value of the small region in the image is divided into subband signals based on the spatial frequency has been described. May be divided into sub-band signals based on. FIG.
In the embodiment shown in FIG.
Although the example in which the number of the delay element 42 used in the PCM and the number of the delay element 47 used in the PCM data restoration unit 45 are one has been described, the number of these delay elements may be plural. Thereby, the prediction error can be reduced.

[0052]

As described above, according to the first or fifth aspect of the present invention, the same signal as the signal of the entire frequency band obtained when the data decompression device performs band synthesis is processed by the data compression device. Then, an error signal between this signal and the original digital signal input to the data compression device is calculated, and the compressed data of the error signal is output together with the compressed data of the signal of each frequency band. As a result, the compressed data of these error signals and the compressed data of the signals of each frequency band are restored to the original error signal and signals of all frequency bands on the data decompression device side, and the restored signals of all frequency bands are restored. If the error signal after the restoration is reduced from, the error can be removed from the signal of the entire frequency band after the restoration, and
The original signal input to the data compression device can be completely restored on the data decompression device side.

Further, the input digital signal is orthogonally transformed and divided into a plurality of frequency band signals, and then the quantization processing is performed on the signal values of each divided frequency band. , It is possible to concentrate signal generation on a specific frequency band. A good data compression ratio is obtained by performing data compression processing such as Huffman coding on the data having a bias in the signal appearance probabilities due to this frequency band, using the high and low signal appearance probabilities of each frequency band. Can be obtained. As a result, unlike conventional data compression devices of the lossless encoding method, sufficient cost-effectiveness can be obtained even if it is necessary to add dedicated devices such as an encoding device and a decoding device to data compression / decompression. The effect can be obtained.

According to the second or sixth aspect of the present invention, a current digital signal is predicted from a past digital signal, and a prediction signal which is a signal of a difference between the predicted digital signal and an actual digital signal is obtained. By calculating the error signal and performing data compression on the prediction error signal, if the digital signal to be compressed is a signal such as a music signal that can be easily predicted, the data of the digital signal is directly compressed. A better data compression ratio can be obtained as compared with the case where the data compression is performed.

According to the third aspect of the present invention, the digital signal compressed by the data compression device is converted into an audio signal, so that the audio signal data obtained after the band division is concentrated on a specific frequency band. It can be data generated by a signal. As a result, by compressing the data of the audio signal after band division in which the appearance probability of the signal is biased by this frequency band, the data compression ratio of the audio signal is compressed. Can be increased.

According to the fourth aspect of the present invention, the digital signal compressed by the data compression device is used as a signal such as luminance and color difference in an image divided into small areas, and the spatial frequency is determined based on these signals. By generating a signal of the spatial frequency and dividing the signal of the spatial frequency into a signal of a plurality of frequency bands, a signal such as luminance and color difference (hereinafter referred to as an image signal) in the image, which often changes smoothly, is generated. By utilizing the property, the data of the image signal after the band division can be converted into data in which signal generation is concentrated in a low frequency band. Thus, by compressing the data of the image signal after the band division in which the appearance probability of the signal is biased by the frequency band, using the bias of the appearance probability of the signal, the data compression ratio of the image signal is reduced. Can be increased.

According to the seventh aspect of the present invention, by causing a computer to read a data compression program and executing the program on the computer, it is possible to obtain the same effect as the above-described invention. it can.

According to the eighth or ninth aspect of the present invention, the signals of each frequency band obtained by decoding the signal output from the data compression apparatus according to the first aspect are synthesized to obtain a full-frequency signal. Band signal is generated, and an error signal obtained by decoding the signal output from the data compression device is subtracted from the signal in the entire frequency band. Digital signal can be completely restored.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of a data compression device according to an embodiment of the present invention.

FIG. 2 is a schematic block diagram of a data decompression device according to an embodiment of the present invention.

3A is a diagram showing an input waveform to a band division filter bank of the data compression device, and FIGS. 3B, 3C, and 3D. FIG.
(E) is a diagram showing waveforms of the sub-band signals S ₀ , S ₁ , S ₂ , and S ₃ output from the band division filter banks.

FIG. 4A is a diagram showing a configuration of a four-output band division filter bank, which is an example of a band division filter bank of the data compression device. FIG. , 2 included in the input waveform
The figure which shows two line spectra corresponding to the component of one frequency.

FIG. 5 is a diagram showing a coordinate system of a monochrome image.

FIG. 6 is a view showing a state in which the monochrome image in FIG. 5 is divided into rectangular small areas.

FIG. 7A is a diagram showing a signal of a luminance value in a small area,
FIG. 2B is a diagram showing a state in which the signal of the luminance value of FIG. 1A is converted into a sub-band signal based on a spatial frequency by a band division filter bank.

FIG. 8 is a diagram illustrating a state in which a small area is scanned in a zigzag manner and signals of luminance values of respective pixels in each small area are sequentially read.

FIG. 9 is a diagram showing a prediction filter used in a preceding stage of the data compression device.

FIG. 10 is a diagram showing a prediction filter used in a subsequent stage of the data decompression device.

FIG. 11 is a block diagram of an embodiment in which the data compression device is used for compressing audio signal data.

FIG. 12 is a block diagram of an embodiment in which the data decompression device is used for restoring audio signal data.

FIG. 13 is a diagram showing a part of a result obtained by compressing music on a commercially available music CD by realizing the data compression device by software.

FIG. 14 is a diagram showing a distribution of the intensity of a subband signal at a certain point in time of a song having a poor compression ratio when compressed using the data compression device.

FIG. 15 is a diagram showing a distribution of the intensity of a subband signal at a certain point in time of a tune having a good compression ratio when compressed using the data compression device.

FIG. 16 is a block diagram of an embodiment in which the data compression device is used for data compression of an image signal.

FIG. 17 is a block diagram of an embodiment in which the data decompression device is used for restoring data of an image signal.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Data compression apparatus 3 Band division filter bank (band division means) 4 Quantization apparatus (quantization means) 5 Subband signal encoding apparatus (first variable length encoding means) 7 Band synthesis filter bank (band synthesis means) Reference Signs List 8 subtractor (error signal calculating means) 9 error signal coding apparatus (second variable length coding means) 21 data decompression apparatus 23 subband signal decoding apparatus (first decoding means) 24 band synthesis filter bank (band synthesis) Means) 25 error signal decoding device (second decoding means) 26 subtractor (error removing means) 41 prediction filter (prediction means) 43 subtractor (prediction error signal calculation means)

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] January 17, 2001 (2001.1.1)
7)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0043

[Correction method] Change

[Correction contents]

Next, the data compression device 1 (FIG. 11) and the data decompression device 21 (FIG. 12) described above with reference to FIG. 13 are realized by software on a personal computer, and music on a commercially available music CD is reproduced. The compression result will be described. A recording medium such as a CD-ROM in which the data compression software program is recorded in this case corresponds to a computer-readable recording medium in which the data compression program is recorded. The compression ratio of each music in the figure is obtained by the following equation (4). Compression rate = data capacity after compression / data capacity before compression × 100 (4) Furthermore, the average compression rate in the figure is obtained by simply averaging the compression rates of all the music pieces in the figure. . The sampling frequency of PCM data was set to 44.1 kHz for all songs.
The data compression device 1 employs a two-channel stereo system, and sets the number of quantization bits for each of the left and right channels to 16 bits. The compression and decompression are performed independently for the left channel input signal and the right channel input signal, and the correlation between the left and right channel signals is not used for compression. As shown in the figure, there is a difference in the data compression ratio depending on the music. However, on average, the data compression ratio is about 57%. When compression was attempted for dozens of songs in various genres, MEGALITH in the figure was the worst compression ratio, and NADIE was the best compression ratio.

[Procedure amendment 2]

[Document name to be amended] Drawing

[Correction target item name] FIG.

[Correction method] Change

[Correction contents]

FIG. 13

Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat II (reference) H04N 7/30 H04N 7/133 Z 7/32 7/137 Z F term (reference) 5C059 KK00 LA00 LB01 MA17 MA23 MA41 MC11 ME02 PP16 SS20 SS30 TA41 TC03 UA02 UA05 UA11 5D045 CC02 5J064 AA01 AA02 BA04 BA09 BA16 BB01 BC06 BC08 BC11 BC15 BC16 BC18 BC19 BC28 BC29 BD04

Claims (9)

[Claims] 1. A band dividing means for orthogonally transforming an input digital signal and dividing the digital signal into a plurality of frequency band signals, wherein a signal value of each frequency band divided by the band dividing means is a predetermined digit or less. Quantizing means for performing a quantization process for rounding the value of the frequency band, while changing the code length of each signal according to the appearance probability of the signal of each frequency band after the quantization process by the quantization means, And a first variable-length encoding unit that compresses the data amount of the signal of the entire frequency band by encoding the signal of A band combining unit that combines signals of the bands to generate a signal of the entire frequency band; a signal of the entire frequency band generated by the band combining unit; and input to the band dividing unit. Error signal calculation means for calculating an error signal that is a signal of a difference from the original digital signal, and a code length of each signal value according to an appearance probability of a value of the error signal calculated by the error signal calculation means. A second variable-length encoding unit that compresses the data amount of the error signal by encoding the error signal while changing the frequency band. Each of the frequency bands compressed by the first variable-length encoding unit is provided. And an error signal data compressed by the second variable-length encoding means. 2. A prediction means for predicting a current digital signal from a past digital signal, and a prediction error for calculating a prediction error signal which is a signal of a difference between the digital signal predicted by the prediction means and an actual digital signal. The data according to claim 1, further comprising a signal calculation unit, wherein the prediction error signal calculated by the prediction error signal calculation unit is input to the band division unit and the error signal calculation unit. Compression device. 3. The data compression apparatus according to claim 1, wherein the digital signal input to said band dividing means and said error signal calculating means is a voice signal. 4. The digital signal input to the band dividing means and the error signal calculating means is a signal such as a luminance and a color difference in an image divided into small areas. 2. The data compression apparatus according to claim 1, wherein a signal of a spatial frequency is generated based on the data, and the signal of the spatial frequency is divided into signals of a plurality of frequency bands. 5. A method for compressing the data amount of an input digital signal, comprising the steps of: orthogonally transforming the input digital signal to divide the signal into a plurality of frequency band signals; For the signal value of the band, performing a quantization process of rounding a value of a predetermined digit or less, while changing the code length of each signal according to the appearance probability of the signal of each frequency band after the quantization process ,
Encoding the signals of the respective frequency bands to compress the data amount of the signals of the entire frequency band; and synthesizing the signals of the respective frequency bands after the quantization processing to generate the signals of the entire frequency band. Calculating an error signal that is a signal of a difference between the generated signal of the entire frequency band and a digital signal that is a transformation source of the orthogonal transformation. Appearance of the value of the calculated error signal According to the probability, while changing the code length of each signal value, by encoding the error signal, compressing the data amount of the error signal, and the data of the compressed signal of each frequency band and Outputting the data of the compressed error signal. 6. The method according to claim 1, further comprising: predicting a current digital signal from a past digital signal; and calculating a prediction error signal that is a signal representing a difference between the predicted digital signal and an actual digital signal. 6. The data compression method according to claim 5, wherein the prediction error signal calculated in the calculating step is a digital signal serving as a source of orthogonal transform in the dividing step. 7. A computer-readable recording medium having recorded thereon a program for compressing the data amount of a digital signal by a computer, wherein the program is a data compression device according to claim 5 or 6. A computer-readable recording medium recording a data compression program, wherein the program is a data compression program for executing the method. 8. A data decompression device for decompressing data compressed by the data compression device according to claim 1, wherein the encoded signal of each frequency band is decoded into an original signal of each frequency band. A first decoding unit that synthesizes the signals of the respective frequency bands after decoding by the first decoding unit to generate a signal of the entire frequency band; A second decoding unit for decoding into an error signal; and an error removing unit for subtracting the error signal decoded by the second decoding unit from the signal of the entire frequency band generated by the band synthesizing unit. Characteristic data decompression device. 9. A data decompression method for decompressing data compressed by the data compression method according to claim 5, wherein the encoded signal in each frequency band is decoded into an original signal in each frequency band. Combining the signals of the respective frequency bands after the decoding to generate signals of the entire frequency band; decoding the encoded error signal into the original error signal; and Subtracting the decoded error signal from the generated signals of all frequency bands.