JP3525614B2 - Digital signal processing device and processing method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital signal processing apparatus and a processing method adopting subband coding for compressing and coding image information, audio information and the like.

[0002]

2. Description of the Related Art Subband coding is known as one of compression coding of audio information and image information. FIG. 13 shows a configuration example of a conventional subband encoding, for example, two-dimensional subband encoding encoder and decoder. A digital image signal is supplied to the input terminal shown by 10. The high-pass filter 11 and the low-pass filter 12 are decomposition filters that divide the horizontal frequency band of the input signal into two. The output signal of the high-pass filter 11 is supplied to the down-sampling circuit 21 that thins out 1/2, and the output signal of the low-pass filter 12 is supplied to the down-sampling circuit 22 that thins out 1/2.

The output signal of the down sampling circuit 21 is supplied to the high pass filter 31 and the low pass filter 32, and the output signal of the down sampling circuit 22 is supplied to the high pass filter 33 and the low pass filter 34. These filters 31 and 32 are decomposition filters that divide the band in the vertical direction into two, and filters 33 and 34 are decomposition filters that divide the band in the vertical direction into two. The respective output signals of the filters 31 to 34 are down-sampling circuits 41, 42, and 4.
3 and 44 are supplied.

With the configuration of FIG. 13, band division as shown in FIG. 14 is performed. Since the high-pass filter 11 separates the high-frequency components of the horizontal frequency and the high-pass filter 31 separates the high-frequency components of the vertical frequency, the output signal of the downsampling circuit 41 is (horizontal high-frequency and vertical high-frequency sub-signals). Band) (denoted as (HH)) signal. Further, the output signal of the down sampling circuit 42 connected to the low pass filter 32 is HL
(Horizontal high frequency band, vertical low frequency band) signal. The low-pass filter 12 separates low frequency components of the horizontal frequency. Therefore, the output signal of the downsampling circuit 43 is an LH (horizontal low band, vertical high band subband) signal, and the output signal of the downsampling circuit 44 is LL.
(Horizontal low band, vertical low band sub-band) signal.

Although not shown in FIG. 13, the output signals of the downsampling circuits 41 to 44 are supplied to the encoding circuit, and the signals of the respective subbands are encoded. Predictive coding, transform coding (for example, DCT), entropy coding, or the like is used as this coding. Efficient coding is applied to the signal of each subband formed by band division, and the transmission image signal is compressed.

The above-mentioned encoder-side output signal is transmitted and supplied to the decoder side. Although not shown, a decoding circuit corresponding to the encoder circuit of each subband on the encoder side is provided, and decoding such as predictive coding and transform coding is performed. The decoded data of each subband is supplied to upsampling circuits 131, 132, 133, and 134 that double the data rate by interpolating 0 data. Upsampling circuits 131, 132, 13
The output signals of 3 and 134 are high-pass filter 1 respectively.
41, low-pass filter 142, high-pass filter 14
3 and the low-pass filter 144. These filters are band synthesis filters.

The output signal of the high pass filter 141 and the output signal of the low pass filter 142 are added by the adder 151. The up-sampling circuit 161 doubles the data rate of the output signal of the adder 151. The output signal of the upsampling circuit 161 is supplied to the high pass filter 171. The adder 152 outputs the output signal of the high-pass filter 143 and the low-pass filter 1
The output signals of 44 are added. The output signal of the adder 152 is supplied to the upsampling circuit 162.

The output signal of the upsampling circuit 161 is supplied to the high-pass filter 171, and the output signal of the upsampling circuit 162 is supplied to the low-pass filter 172. The output signal of the high pass filter 171 and the output signal of the low pass filter 172 are supplied to the adder 181. The decoded image signal is extracted from the adder 181 to the output terminal 190.

[0009]

In the system using the conventional subband coding, when the predictive coding is used as the coding of each subband, the accuracy of the predicted value affects the compression efficiency. Conventionally, the average value of the decoded values of pixels in the vicinity of the pixel to be predicted has been used as the prediction value. As a result, the accuracy of prediction was insufficient.

Therefore, an object of the present invention is to provide a digital signal processing apparatus and a processing method capable of further improving the compression efficiency by improving the accuracy of a prediction value when a subband signal is encoded by predictive encoding. To provide.

[0011]

SUMMARY OF THE INVENTION The present invention, in the digital signal processing apparatus designed to compress the data by sub-band coding, a filter means for dividing the input digital signal a plurality of sub-bands, the number of double Downsample for downsampling subband digital signals
Ring means and a downsampled first subband digital signal from the downsampled first subband digital signal .
Prediction means for predicting digital signal of 2 subbands
And, by using the prediction value from the prediction means, down San pre
The compressed second subband digital signal
And a predictive coding means, the predicting means being the first sub-bar.
Classes classified according to the characteristics of the digital signal
The first corresponding to the digital signal of the first subband for each
Learning digital signal and the second subband digit
Between the second learning digital signal corresponding to the digital signal
Based on the relationship between subbands obtained by learning
1st subband digital signal to 2nd subband
The digital signal processing device is characterized by predicting the digital signal of . Further, the present invention is a digital signal processing method so as to compress the data by sub-band coding, dividing step of dividing the input digital signal of a plurality of sub-bands, downsampling the digital signals of a plurality of sub-bands Down sample
And ring step, which is down-sampled from the digital signal of the first sub-bands downsampled
A prediction step for predicting the digital signal of the second subband.
Up and down using the predicted value from the prediction step.
Sampled second subband digital signal
Compressing the first
Classified based on the characteristics of subband digital signals
Corresponds to the first sub-band digital signal for each class
Of the first learning digital signal and the second sub-band
Second learning digital signal corresponding to the digital signal
Based on the relationship between the subbands obtained by learning between
From the digital signal of the first subband to the second subband.
A digital signal processing method characterized by predicting a band digital signal .

In the sub-band coding, when the signals of the sub-bands other than the horizontal low-band and vertical high-band sub-bands LL are compressed by the predictive coding, the predictive value is generated by the class classification adaptive process, so that the prediction accuracy is high. It is possible to improve the compression efficiency.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows an encoder system according to an embodiment of the present invention. The digital image signal from the input terminal 10 is supplied to the high pass filter 11 and the low pass filter 12, and the horizontal frequency band is divided into two. The output signal of the high pass filter 11 is supplied to the down sampling circuit 21, and the output signal of the low pass filter 12 is supplied to the down sampling circuit 22. The downsampling circuits 21 and 22 are for thinning out pixels by subsampling to reduce the signal rate to 1/2.

The output signal of the down sampling circuit 21 is supplied to the high pass filter 31 and the low pass filter 32. The vertical frequency band is divided into two by the high pass filter 31 and the low pass filter 32. Similarly, the output signal of the down-sampling circuit 22 is divided into two vertical frequency bands by the high-pass filter 33 and the low-pass filter 34. These filters 31-
Down sampling circuit 41,
42, 43 and 44 are connected.

The filters 11, 12, 41-44 described above
Operates as a decomposition filter, and the signal band is divided into four subbands. That is, similarly to the configuration on the encoder side of the related art described above, the output signal of the down sampling circuit 41 is an HH (horizontal high range, vertical high range) signal, and the output signal of the down sampling circuit 42 is HL (horizontal high range,
Vertical down band), the output signal of the down sampling circuit 43 is LH (horizontal low band, vertical high band) signal, and the output signal of the down sampling circuit 44 is LL (horizontal low band, vertical low band) signal. is there.

The HH signal from the down sampling circuit 41 is supplied to the adder 51. The prediction signal HH ′ is inverted and supplied from the prediction unit 61 to the adder 51. Therefore, the adder 51 generates the difference between the input HH signal and the prediction signal HH '. The output signal of the downsampling circuit 42 is supplied to the adder 52. The adder 52 includes a prediction unit 6
The output signal 2 is inverted and supplied, and the adder 62 generates a difference between the HL signal and the HL ′ signal. further,
The output signal of the down sampling circuit 43 is supplied to the adder 53. The prediction signal LH ′ is inverted and supplied from the prediction unit 63 to the adder 53, and LH ′ is supplied from the adder 53.
And LH '.

The output signal (difference signal) of the adder 51 is supplied to a compression code encoder 81, and the output signal is taken out to an output terminal 91o as a subband (HH) coded output signal. The output signal of the encoder 82 is the subband (H
L) is output as an encoded output signal to the output terminal 92o. The output signal of the encoder 83 is taken out to the output terminal 93o as a subband (LH) encoded output signal.
Furthermore, the sub-band (L
The image signal of L) is supplied to the encoder 84. The encoder 84 performs compression encoding, and the output terminal 94o
, The subband (LL) encoded output signal is extracted.

The decoder 71 is connected to the encoder 84. The decoder 71 decodes the output signal of the encoder 84. The output signal of the decoder 71 (that is, the decoded subband (LL) signal) is supplied to the prediction units 61, 62, and 63, respectively. The prediction units 61, 62, 63
As will be described later, the sub-bands (HH, H
L, LH) signals LH ′, HL ′, HH ′ are predicted.

The encoders 81, 82, 83, 84 perform compression encoding based on the characteristics of the signals in each subband. Since the sub-band (LL) has a signal characteristic similar to that of the input image signal, the encoder 84 uses the DCT (Disc
rete Cosine Transform), ADRC (Adaptive Dynamicr)
The data is compressed by compression encoding such as an ange coding, vector quantization, and DPCM. The encoders 81, 82, and 83 encode the difference signal, and thus perform data compression by entropy encoding such as Huffman code.

The encoded output of each sub-band on the encoder side described above is transmitted via a communication path or recorded on a recording medium. The encoded signals received or reproduced are input terminals 91i, 92i, 93i, 94 of the decoder shown in FIG.
i respectively. The subband (HH) encoded signal from the input terminal 91i is supplied to the decoder 101. Further, the subband (HL) encoded signal from the input terminal 92i is supplied to the decoder 102, and the input terminal 93i
The subband (LH) encoded signal from the decoder 103
Subband (LL) from the input terminal 94i
The encoded signal of is supplied to the decoder 104.

Decoders 101, 102, 103 are provided corresponding to encoders 81, 82, 83 on the encoder side and perform, for example, entropy coding decoding. Decoder 1
04 is provided corresponding to the encoder 84. The decoded signal obtained from each of the decoders 101, 102, 103 is a difference signal, and this signal is added by the adders 111, 11
2, 113 and 114, respectively. The adder 111, 112, 113, 114 includes a prediction unit 12
Predictive signals from 1, 122, 123 are provided.

The predicting units 121, 122, 123 perform the class classification adaptive processing similarly to the predicting units 61, 62, 63 on the encoder side, based on the decoded signal of the subband LL from the decoder 104. HL, LH) prediction signals are generated. Therefore, the decoded signal of the subband HH is obtained from the adder 111. Adder 112, 1
Subband (HL, LH) decoded signals are obtained from 13 respectively. Up-sampling circuits 131, 132, for each of the adders 111, 112, 113,
133 is connected, and the upsampling circuit 134 is connected to the decoder 104. For example, by interpolating 0 data, the upsampling circuits 131 to 1
34 doubles the sampling rate.

A high pass filter 141 for band synthesis, a low pass filter 142, a high pass filter 143 and a low pass filter 144 are connected to the upsampling circuits 131 to 134. High pass filter 1
The output signals of 41 and the low-pass filter 142 are added by the adder 151, and the output signals of the high-pass filter 143 and the low-pass filter 144 are added by the adder 152. The output signals of the adders 151 and 152 are supplied to the upsampling circuits 161 and 162, respectively, and the sampling rate is doubled. Then, the high-pass filter 171, the low-pass filter 172, and the adder 181 perform band combination. Adder 18
The decoded signal is taken out from 1 to the output terminal 190. The configuration and processing after the up-sampling circuits 131 to 134 are the same as the sampling rate change and band synthesis performed in the conventional sub-band encoding decoder shown in FIG.

Next, another embodiment of the present invention will be described. FIG. 3 shows the configuration of the encoder side of another embodiment, and FIG. 4 shows the configuration of the decoder side thereof. The other embodiment is different from the first embodiment in that when a signal in a subband other than the subband LL is predictively encoded, prediction is performed in multiple stages over a plurality of subbands. On the encoder side shown in FIG. 3, the input digital image signal is horizontally divided into two bands by a high-pass filter 11 and a low-pass filter 12, and a high-pass filter 31,
The band is vertically divided into two by the 33 and the low-pass filters 32 and 34. In addition, the down sampling circuit 2
The sampling rate is set to 1/4 by 1, 22, 41, ..., 44. These processes are the same as in the above-described embodiment.

The adder 52 is supplied with the signal of the sub-band HL and the inverted signal of the prediction signal HL 'from the prediction unit 62, and a difference signal is generated at its output. The adder 53 is supplied with the signal of the subband LH and the inverted signal of the prediction signal LH ′ from the prediction unit 63, and a difference signal is generated at its output. The prediction units 62 and 63 generate prediction signals HL ′ and LH ′ by the class classification adaptive processing based on the decoded signal of the subband LL from the decoder 73.

The process of predictively coding the signal of the subband HH uses the decoded signal of the coded signal of the subband HL (or LH). That is, the decoded signal of the subband HL is supplied to the prediction unit 61, and the prediction signal HH ′ is formed. The decoded signal of the subband HL can be generated by adding the predicted signal HL ′ from the prediction unit 62 and the decoded difference signal from the decoder 72 by the adder 68.

FIG. 4 shows the configuration of the decoder side according to another embodiment of the present invention. Predictors 122 and 123 generate prediction signals HL ′ and LH ′, respectively, based on the decoded signal of subband LL from decoder 104, and adder 112
And 113, the decoded signals of subbands HL and LH are obtained. Also, the subband HL from the adder 112
The decoded signal of HH 'is supplied to the prediction unit 121.
Is generated. In the adder 111, the difference signal from the decoder 101 and the prediction signal HH ′ are added. A decoded signal of the subband HH is obtained at the output of the adder 111.

In the above-described embodiment, the signals of all the other subbands are predicted from the decoded signal of the subband LL. On the other hand, in another embodiment, the sub-band HL (or L
The signal of subband HH is predicted based on the decoded signal of H). Subband HL compared to subband LL
Since (or LH) is a signal closer to the subband HH, the prediction accuracy can be further improved. As a result, it is possible to further improve the compression rate of the subband HH.

Still another embodiment of the present invention will be described. As shown in FIG. 5, the sub-band LL of the digital image signal is further divided into two in terms of horizontal and vertical frequencies. As a result, in the subband LL, the horizontal low range,
A vertical low band sub-band LLLL, a horizontal high band, a vertical low band sub band LLHL, a horizontal low band, a vertical high band sub band LLLH, and a horizontal high band, a vertical high band sub band LLHH are formed.

Such band division is performed by the configuration shown in FIG. The signal of the sub-band LL that has passed through the down-sampling circuit (not shown) is the high-pass filter 13.
And a low-pass filter 14 to divide the band with respect to the horizontal frequency. Down sampling circuits 23 and 24 are connected to these filters. The horizontal frequency high frequency signal separated by the high pass filter 13 is the high pass filter 35 and the low pass filter 36.
, And the band is divided with respect to the vertical frequency. The horizontal frequency low-frequency signal separated by the low-pass filter 14 is the high-pass filter 37 and the low-pass filter 38.
, And the band is divided with respect to the vertical frequency. Down-sampling circuit 45 for these filters,
46, 47 and 48 are connected. Therefore, the sub-band L is output to each output of the down-sampling circuits 45 to 48.
The LHH, LLHL, LLLH, and LLLL signals are obtained separately.

In yet another embodiment of the present invention, predictive coding is applied to newly formed subbands other than subband LLLL. FIG. 7 shows a configuration for predictive coding according to still another embodiment. Adders 51, 52, 5
Signals of subbands HH, HL, and LH are supplied to the signal line 3. The signals of the divided subbands LLHH, LLHL, and LLLH are supplied to the adders 55, 56, and 57. Adders 51, 52, 53, 55,
56 and 57 include predictors 61, 62, 63, 65, 6
The prediction signal formed by 6, 67 is provided. These prediction units generate prediction signals by the class classification adaptive processing. The horizontal low band and vertical low band signals LLLL in the subband LL are encoded by the encoder 88. The encoded signal is taken out at the output terminal 98o.

As a signal on which the prediction signal is based, a signal whose frequency is as close as possible to the signal to be predicted is used. Here, the prediction signal H of the subband HH
H'is formed based on the decoded signal of the subband LLHH. The coded signal of the encoder 85 taken out to the output terminal 95o is decoded by the decoder 74, and the decoded output and the prediction signal LLHH ′ are added by the adder 78 to form a signal of the subband LLHH. ing.

The prediction signal HL 'of the subband HL is formed based on the decoded signal of the subband LLHL. The coded signal of the encoder 86 taken out to the output terminal 96o is decoded by the decoder 75, and the decoded output and the prediction signal LLHL ′ are added by the adder 79 to form a signal of the subband LLHL. ing. Subband L
The H prediction signal LH ′ is formed based on the decoded signal of the subband LLLH. The decoder 76 decodes the coded signal of the encoder 87 taken out to the output terminal 97o,
Further, the decoded output and the prediction signal LLLH 'are added by the adder 80.
The signal of the sub-band LLLH is formed by adding.

Prediction signal LLHH 'of subband LLHH
Is formed based on the decoded signal of the subband LLHL obtained from the adder 79, similarly to the prediction signal of the subband HL described above. Prediction signal LL of subband LLHL
HL 'is formed based on the decoded signal of the subband LLLL. By decoding the encoded signal taken out at the output terminal 98o by the decoder 77, the subband L
It forms the LLL signal. The prediction signal LLLH ′ of the subband LLLH is also formed based on the decoded signal of the subband LLLL.

FIG. 8 shows the configuration of the decoder side according to still another embodiment of the present invention. However, in FIG. 8, for simplification, a circuit for upsampling each signal of the subbands HH, HL, LH, and LL, and a signal for synthesizing the signals of these subbands to form a decoded image signal The filter is omitted.

The encoded signals obtained at the output terminals 91o to 98o on the encoder side are supplied to the input terminals 91i to 98i via a transmission line or a recording medium. For these input terminals 91i to 98i, encoders 81 to 88 on the encoder side and corresponding decoders 101 to 10
8 are connected. From the decoders 101 to 107, decoded signals (difference signals) for predictive coding of each subband are output,
The decoder 108 outputs a subband LLLL signal.

Adders 111 to 117 are connected to the decoders 101 to 107, respectively. Adders 111 to 11
7 is a difference signal from the decoders 101 to 107 and a prediction unit 12
The prediction signals generated by 1 to 127 are added to generate decoded signals for each subband. Prediction units 121 to 12
7 has the same relationship as the encoder side, and generates a prediction signal of another subband based on a decoded signal of another subband.

That is, the prediction signal HH 'is added to the adder 11
5 is formed on the basis of the decoded signal of the subband LLHH from 5, the prediction signal HL ′ is formed on the basis of the decoded signal of the subband LLHL from the adder 116, and the prediction signal L
H ′ is formed on the basis of the decoded signal of the subband LLLH from the adder 117, and the prediction signal LLHH ′ is formed on the basis of the decoded signal of the subband LLHL from the adder 116, and the prediction signals LLHL ′ and LLLH ´ is
It is formed based on the decoded signal of the subband LLLL from the decoder 108.

The decoded signals of the subbands HH, HL and LH are obtained from the adders 111, 112 and 113, respectively.
Subband LLH from adders 115, 116, 117
The decoded signals of H, LLHL, and LLLH are obtained. Decoded subbands LLHH, LLHL, LLL
The H and LLLL signals are upsampling circuits 135,
136, 137 and 138 respectively.
Then, the high pass filters 145, 147, the low pass filters 146, 148, the adders 153, 154, the upsampling circuits 163, 164, the high pass filter 1
Band combining is performed by 73, 174 and the adder 175, and a decoded signal of the subband LL is obtained.

In yet another embodiment of the present invention described above, the compression efficiency can be improved by further dividing the subband LL having both low horizontal and vertical frequencies into a plurality of subbands.

According to the present invention, when predictive-encoding a signal in a subband other than the subband having the lowest horizontal and vertical frequencies, a predictive signal is generated by class classification adaptive processing. The class classification adaptive processing is to classify the input signal into several classes based on the characteristics of the input signal, and execute an appropriate adaptive processing for each class prepared in advance.

Generation of a prediction signal to which the class classification adaptive process is applied will be described below with reference to the drawings. When a prediction signal LH ′ of another subband such as LH is generated from a decoded signal of a certain subband such as LL, the classification is performed on the decoded signal of the subband LL (for example, 8-bit PCM).
A class generation tap is set for (data) and a class is generated according to the waveform characteristics of this signal. The following has been proposed as a method for generating a class of signal waveforms.

1) Use PCM data directly. 2) Apply ADRC (Adaptive Dynamic Range Coding) to reduce the number of classes. 3) Apply DPCM (Predictive Coding) to reduce the number of classes. 4) Apply VQ (Vector Quantization) to reduce the number of classes. 5) Perform class classification in the frequency domain such as DCT (discrete cosine transform).

For example, when 7 pieces of PCM data are directly used, the number of classes becomes a huge number of 2 ⁵⁶ because there are 7 pixels of 8-bit data, which is a problem in practical use. Therefore, in practice, ADRC (encoding adapted to the dynamic range) or the like is applied to reduce the number of classes. For example, when 1-bit ADRC is applied to 7-tap data, the minimum value of 7 pixels is removed based on the dynamic range defined by the 7-pixel data, and then the pixel value of each tap is adaptively 1-bit quantized. So do 12
It is possible to reduce to 8 classes.

ADRC was developed as a signal compression method for VTR, but is suitable for expressing the waveform characteristics of an input signal with a small number of classes. Further, as a method of classifying, classifying may be performed based on the activity of the input signal. For example, the classification may be performed based on the difference (dynamic range) between the maximum value and the minimum value of the block of a plurality of pixels of the input signal. Furthermore, the class classification based on the activity and the class classification based on the waveform characteristics of the input signal described above may be used together. The adaptive process is executed for each class thus classified, and an example thereof is a prediction process using a prediction coefficient for each class generated by learning in advance. An example of the prediction formula is shown in formula (1).

[0046]

[Equation 1]

Y ': estimated pixel value x _i of subband LH: pixel value w _i of signal prediction of subband LL: prediction coefficient

The pixel value of the subband LH is estimated by the product-sum operation of the prediction coefficient thus generated for each class and the input data. FIG. 9 shows an example of the circuit configuration of the class classification adaptive processing. An input signal (signal of the subband LL) is supplied from the input terminal 201, and the input signal is supplied to the class classification unit 202 and the prediction calculation unit 204. The class classification unit 202 generates a class for the input signal based on the class classification processing as described above. The generated class is supplied to the prediction coefficient ROM 203 as an address, and the prediction coefficient ROM 203 outputs the class to the prediction calculation unit 204.
The prediction coefficient is output to. In the prediction calculation unit 204,
The prediction calculation of Expression (1) is executed using the input signal and the prediction coefficient, and the output (prediction signal LH ′) is obtained at the output terminal 205.

The above-mentioned prediction coefficient is generated by learning in advance. The learning method will be described below. Formula (1)
An example in which a prediction coefficient based on the linear first-order combination model of is generated by the least square method will be described. The least squares method is applied as follows. As a generalized example, X is input data, W
Consider the following equation, where is a prediction coefficient and Y is an estimated value.

Observation equation; XW = Y (2)

[0051]

[Equation 2]

The least squares method is applied to the data collected by the above observation equation. In the example of formula (1), n
= 13, m is the number of learning data. Consider the residual equation of equation (4) based on the observation equations of equations (2) and (3).

[0053]

[Equation 3]

From the residual equation of the equation (4), it is considered that the most probable value of each w _i is a condition that the sum of squared errors is minimized. The sum of squared errors is given by the following equation.

[0055]

[Equation 4]

That is, it is sufficient to consider the condition of the following equation (5).

[0057]

[Equation 5]

Considering n conditions based on i in equation (5),
It is sufficient to calculate w ₁ , w ₂ , ..., W _n satisfying this. Then, the equation (6) is obtained from the residual equation (4).

[0059]

[Equation 6]

Equation (7) is obtained from Equations (5) and (6).

[0061]

[Equation 7]

Then, the normal equation (8) is obtained from the equations (4) and (7).

[0063]

[Equation 8]

In the normal equation of the equation (8), since it is possible to set up the same number of equations as the unknown number n, each w _i
The most probable value of can be obtained. Then, the sweeping method (Gauss-Jordan elimination method) is used to solve the simultaneous equations.

Here, an example of learning the prediction coefficient by software using the above least square method is shown in the flowchart of FIG. First, this flow chart starts with the formation of learning data in step S1, and in this step S1, input data (for example, a signal of subband LL) and a teacher signal to be predicted (for example, a signal of subband LH)
To prepare. Then, in the class determination in step S3, the input data is classified into classes.

In step S4, the normal equation of equation (8) is generated for each class from the image value of the prediction tap formed from the input signal and the value of the teacher signal. In general,
In order to eliminate the influence of noise, we will exclude those with small activity of input data change from the learning target. In this learning process, a normal equation in which a lot of learning data is registered is generated. The normal equation generation process is repeated until the end of the number of learning target data is confirmed in step S2.

When all the target learning data are completed,
The control shifts to the determination of the prediction coefficient in step S5. here,
Formula (8) for each class generated from many learning data
The normal equation of is solved. As the solution method for the simultaneous equations, the above-mentioned sweeping method is used. The prediction coefficient thus obtained is registered in a storage unit such as a ROM that is divided into addresses according to classes in the process of registering the prediction coefficient in step S6. Through the above learning process, the prediction coefficient of the class classification adaptive process is generated.

In addition to the class classification adaptive processing by the above prediction calculation, a class classification adaptive processing using an optimum predicted value generated in advance by the centroid method has also been proposed. Figure 11
As shown in FIG. 3, an input signal (for example, a signal of subband LL) is supplied from the input terminal 201, and the class classification unit 206
Is supplied to. The class classification unit 206 determines the class of the pixel of interest of the subband LH to be predicted using the input signal. This class or optimum predicted value ROM20
7 is supplied as an address. The ROM 207 stores the predicted value obtained by learning in advance for each class. Therefore, the prediction signal LH is output to the output terminal 205.
'Is obtained.

An example of learning the optimum predicted value stored in the ROM 207 will be described with reference to the flowchart shown in FIG. The process shown in this flowchart starts from step S11, and in step S11, the frequency counters N (*) of all classes and the data tables E (*) of all classes are initialized. If a class is C0, the counter of the corresponding frequency is N (C
0), and the corresponding data table is defined as E (C0). Also, * indicates all classes.

In the class detection in step S12, the class C is determined from the learning target pixel neighborhood data. As a method of class classification, in addition to ADRC as described above, expression methods such as PCM expression, DPCM, BTC (block truncation coding), VQ, and orthogonal transform can be considered. Moreover, when considering the activity of the block composed of the data to be classified, the number of classes should be increased by the kind of classification by the activity.

Then, in step S13, the pixel value y to be predicted is detected, and the detected pixel value y is added for each class C in step S14. That is, after adding y to the content of the data table E (C) of the class C, the frequency counter N (C) of the learning pixels of the class C is incremented by 1 in step S15.
In step S16, the above processing is repeatedly executed for all learning target pixels to determine whether the final frequency counters N (*) of all classes and the corresponding data tables E (*) of all classes have been generated. To be judged. Young
When it is determined that all the data have been completed, the control moves to step S17.

In step S17, the average value of each class is calculated by dividing the data integrated value, which is the content of the data table E (*) of each class, by the frequency of the frequency counter N (*) of the corresponding class. calculate. This value is the optimum predicted value for each class by the center of gravity method. The name of the centroid method is derived by taking the average of the distribution of the learning target pixel values.
Finally, in step S18, the learning by the center of gravity method ends by registering the class and the above-described optimum predicted value in the storage means such as the ROM. Further, in order to eliminate the influence of noise in the learning process, pixels with small activity are excluded from the learning target.

The present invention can be applied not only to digital image signals but also to subband coding of digital audio signals. Further, although the linear first-order coupling equation is used when forming the predicted value, the present invention is not limited to this, and a higher-order coupling equation may be used.

[0074]

According to the present invention, when the compression is performed by the predictive coding in the subband coding, the predictive signal is formed by the class classification adaptive processing, so that the accuracy of the prediction can be improved and the compression efficiency can be improved. Can be improved. Further, when the present invention is applied to the encoding of an image signal, it is possible to transmit a plurality of image signals whose resolution gradually decreases from a high resolution image signal, and it is possible to realize multiple resolution. is there.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration on an encoder side of an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a decoder side according to an embodiment of the present invention.

FIG. 3 is a block diagram showing a configuration of an encoder side of another embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a decoder side according to another embodiment of the present invention.

FIG. 5 is a schematic diagram for explaining band division according to still another embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of band division according to still another embodiment of the present invention.

FIG. 7 is a block diagram showing a configuration of an encoder side according to still another embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of a decoder side according to still another embodiment of the present invention.

FIG. 9 is a block diagram showing an example of a prediction unit by a class classification adaptive process.

FIG. 10 is a flowchart illustrating a learning method of prediction coefficients.

FIG. 11 is a block diagram showing another example of a prediction unit by the class classification adaptive processing.

FIG. 12 is a flowchart illustrating a method of learning a predicted value.

FIG. 13 is a block diagram showing a configuration of a conventional encoder side and decoder side of subband encoding.

FIG. 14 is a schematic diagram for explaining band division in subband coding.

[Explanation of symbols]

10 ... Digital image signal input terminals 51 to 53
... Adding circuit for generating difference signal between input signal and prediction signal, 61-63 ... Prediction unit

─────────────────────────────────────────────────── ─── Continuation of the front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) H03M 3/00-11/00 H04N 1/41 H04N 7/32

Claims (13)

(57) [Claims] 1. A digital signal processing apparatus designed to compress the data by sub-band coding, a filter means for dividing the input digital signal of a plurality of sub-bands, the digital signals above Kifuku number of subbands downsampling means for downsampling a second subband downsampled from the digital signal of the first sub-bands downsampled
Prediction means for predicting a digital signal, using the prediction value from the prediction means, down San pre
The compressed digital signal of the second sub-band.
And a predictive coding means for compressing the digital signal of the first subband.
For each class classified based on the characteristics of the
First learning digit corresponding to band digital signal
Tal signal and the digital signal of the second sub-band
Learning with the corresponding second learning digital signal
Based on the relationship between the subbands obtained,
Of the second sub-band from the sub-band digital signal
A digital signal processing device characterized by predicting a digital signal. 2. The digital signal processing device according to claim 1, wherein the predicting means is the predicted digital signal of the second subband.
Based on the
The digital signal of the third subband of the
However, the predictive coding means predicts the predicted digital signal of the third sub-band.
Using the measured value, the predicting means uses the predicted digital signal of the second sub-band.
The above for each class classified by the signal characteristics based on
The second learning corresponding to the digital signal of the second subband.
The conventional digital signal and the third subband digit
Between the third learning digital signal corresponding to the digital signal
Based on the relationship between the subbands obtained by learning in
From the signal based on the second sub-band digital signal
A device for predicting a digital signal of the third subband . 3. The digital signal processing device according to claim 1, wherein when a prediction value is generated, a digital signal of a subband on a lower frequency side is used as the first subband. apparatus. 4. A digital signal processor according to claim 1, said relationship, to form a learning-pair in the upper Symbol first learning digital signal and said second digital signal for learning, a plurality of the and wherein the obtained result of the learning between band component of said first and second learning digital signals corresponding to the sub-band. 5. The digital signal processing device according to claim 1, wherein the predictive coding means is a digitizer of the first subband.
Class classifying means for classifying with a digital signal , storage means for storing prediction coefficients acquired by learning in advance corresponding to the class output from the class classifying means, and storage means corresponding to the class Computation means for computing an optimum prediction value of the digital signal of the second sub-band by computing the prediction coefficient read from the digital signal of the first sub-band. A device characterized by. 6. The digital signal processing device according to claim 5 , wherein the arithmetic means is the prediction coefficient and the first subband.
The second by the linear combination equation of a digital signal
A device for generating a prediction value of a digital signal of the sub-band . 7. The digital signal processing device according to claim 5 , wherein when the prediction coefficient for each class is acquired by learning, a signal having a small activity is excluded from learning targets. 8. The digital signal processing device according to claim 1, wherein the predictive coding means is a digitizer of the first subband.
And class classification means for performing classification by Tal signal, a pre-optimum prediction value of the digital signal of the acquired said second subband by learning, memory means for storing in response to the class output from the class classification means It has the door, in correspondence with the class and wherein the obtaining the optimum predicted value from said storage means. 9. The digital signal processing device according to claim 8 , wherein the optimum predicted value prepared for each class is obtained in advance by a centroid method. 10. The digital signal processing device according to claim 8 , wherein when acquiring the optimum predicted value by learning for each of the classes, a signal having a small activity is excluded from the learning target. 11. The digital signal processing device according to claim 5 or 8 , wherein the predictive coding means includes a digitizer for the first subband.
An apparatus characterized by performing class classification based on the activity of a Tal signal . In the digital signal processing apparatus according to claim 12] according to claim 1 1, the predictive encoding means, daisy of the first sub-band
Classification based on Tal signal activity and above
The first subband of the first plurality of subbands
An apparatus characterized by performing the class classification in combination with the class classification based on the waveform of the digital signal . 13. A digital signal processing method for compressing data by subband coding, wherein a dividing step of dividing an input digital signal into a plurality of subbands, and downsampling the digital signals of the plurality of subbands. Down-sampling step and down-sampled second sub-band from the down-sampled first sub-band digital signal
A prediction step of predicting a digital signal, using the prediction value from the prediction step, down San
Pulled digital signal of the second sub-band
, And the predicting step comprises digitizing the first subband.
For each class classified based on the characteristics of the
The first learning data corresponding to the subband digital signal.
The digital signal and the digital signal of the second sub-band.
Learning with the second learning digital signal corresponding to the No.
Based on the relationship between the subbands obtained by
From the digital signal of the sub-band of
A digital signal processing method characterized by predicting a digital signal of a digital signal.