JPH0671332B2 - Vector quantizer - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vector quantizer that encodes image signals and audio signals with high efficiency.

2. Description of the Related Art Image data has a large amount of information, which increases the cost of the transmission line in image transmission. Therefore, an encoder for encoding an image signal with high efficiency and transmitting at a low transfer rate and a decoder for quickly decoding the transmitted signal have been developed.

FIG. 2 is a block diagram of a conventional vector quantizer.

In FIG. 2, 201 is a signal input terminal, 202 is a distortion calculation circuit, 203 is a waveform codebook ROM, 204 is an address counter, 205 is a minimum distortion detection circuit, 206 is an index latch,
Reference numeral 207 is a signal output terminal.

The operation of the vector quantizer configured as described above will be described below. The block signal input from the signal input terminal 201 is a distortion calculation circuit as an input vector.
Entered in 202. When the block input vector is sent to the distortion calculation circuit 202, the address counter 204 counts up, and the waveform codebook ROM 203
The output vector from is read out sequentially. Distortion calculation circuit 20
Reference numeral 2 calculates the absolute value of the difference between the output vector from the waveform codebook ROM 203 and the input vector from the signal input terminal 201 and each element, calculates the sum, and sends it as a distortion value to the index latch. The minimum distortion detection circuit 205 sends the minimum distortion recognition signal to the index latch 206 when the minimum distortion is detected from the distortion values from the distortion calculation circuit 202 within the number of output vectors stored in the waveform codebook ROM 203. The index latch 206 reads the count number from the address counter 204 when the minimum distortion recognition signal is sent, and the signal output terminal 207 when the count number has counted up the number of output vectors stored in the waveform codebook ROM 203. The count number is sent to (for example, Chapter 5 of "High Efficiency Coding Technology in Image Transmission", published by Trikeps).

However, in the above configuration, the waveform codebook RO applied to the input vector created from various input signals is used.
In addition to the large number of output vectors required in M, the number of distortion calculations increases, which causes a large load on the distortion calculator.

In view of the above problems, the present invention can significantly reduce the number of output vectors by separating the minimum value in the input vector and normalizing the maximum value, and reduce the load on the distortion calculator. .

Means for Solving the Problems In order to solve the above problems, a vector quantizer of the present invention uses an input signal that is an input signal that is blocked in advance for each of k (k is an integer) samples, and A minimum value separation circuit that calculates the minimum value of the elements, outputs the minimum value separation vector on which the minimum value separation processing is performed by subtracting the minimum value from each element of the input vector, and outputs the minimum value, The maximum value of the elements of the minimum value separation vector is calculated, and the maximum value is output together with the minimum value separation normalized vector that is normalized by dividing the maximum value as an amplitude coefficient from each element. And an amplitude normalization circuit that sequentially counts up the count value to a predetermined number each time the minimum value separation normalization vector is read from the amplitude normalization circuit. An output address counter and a plurality of output vectors optimized in advance based on the statistical properties of the vector separated from the minimum value and normalized are stored, and the output vector is sequentially output every time the count value is counted up. And a waveform codebook ROM that outputs a reset signal when the count value is counted up by a predetermined number, and a distortion is generated between the minimum value separation normalization vector and the output vector read from the waveform codebook ROM. The distortion calculation circuit that obtains and outputs the distortion value and the reset signal from the waveform codebook ROM are input to update the distortion value, and the distortion values from the distortion calculation circuit are sequentially compared to give the minimum distortion. A minimum distortion detection circuit that outputs a distortion recognition signal, and when the minimum distortion recognition signal is input, reads the count value from the address counter, An index latch that outputs a read value as an index value when a reset signal from the waveform codebook ROM is input, a minimum value from the minimum value separation circuit, a maximum value from the amplitude normalization circuit, and the index latch It is configured so as to include an encoding circuit for performing the predetermined encoding of the index values from.

Operation The present invention reduces the number of patterns of the input vector by separating the minimum value of the input vector and normalizing the minimum value of the input vector by the above configuration, thereby reducing the number of output vectors in the waveform codebook ROM. Therefore, the strain on the distortion calculator can be reduced.

Embodiment A vector quantizer according to an embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the vector quantizer of the present invention. In FIG. 1, 101 is a signal input terminal, 102 is a minimum value separation circuit, 103 is an amplitude normalization circuit, 104
Is an address counter, 105 is a waveform codebook ROM, 106
Is a distortion calculation circuit, 107 is a minimum distortion detection circuit, 108 is an index latch, 109 is an encoding circuit, and 110 is a signal output terminal.

The operation of the vector quantizer configured as described above will be described below with reference to FIG. A signal S = {S ₁ , S blocked from the signal input terminal 101 every 9 samples
₂ , ..., S ₉ } is input as an input vector. For example, when S ₂ of the element of the input vector S is the minimum value, the minimum value separation circuit 102 obtains the minimum value S ₂ and the minimum value index 2 indicating the position of the element of the input vector to which the minimum value is given, The minimum value separation vector T by subtracting the minimum value S ₂ from each element
Is calculated according to the equation (1).

_{T L = S L -S 2 (} L = 1,2, ......, 9) ...... (1) provided that _{T = {T 1, T 2} , ......, T 9} the minimum value S ₂ of the thus obtained The minimum value index 2 is sent to the encoding circuit 109, the minimum value index 2 is sent to the address counter 104, and the minimum value separation vector T is sent to the amplitude normalizing circuit 103. Now, assuming that T ₇ of the element of the minimum value separation vector T is the maximum value, the amplitude normalization circuit 103 outputs the maximum value T ₇ and the maximum value indicating the position of the element of the minimum value separation vector given the maximum value. The index 7 is obtained, and the maximum value T ₇ is divided from each element to obtain the minimum value separated normalization vector X according to the equation (2).

X _L = T _L / T ₇ (L = 1,2, ..., 9) (2) where X = {X ₁ , X ₂ , ..., T ₉ } The maximum value T ₇ obtained there is The waveform codebook ROM 105 and the encoding circuit 109 are sent, the maximum value index 7 is sent to the address counter 104, and the minimum value separation normalization vector X is sent to the distortion calculation circuit 106.

The first table is a table relating to the count value output by the address counter 104.

When the minimum-value-separated normalized vector X is sent to the distortion calculation circuit 106, the address counter 104 sequentially waveforms the count value that counts up to 72 times with (B) selected in Table 1 as the group number 2. It is sent to the codebook ROM 105.

The second table is a table relating to the control values in the waveform codebook ROM 105 determined by the maximum value from the amplitude normalization circuit 103.

Now, assuming that the maximum value T ₇ from the amplitude normalization circuit 103 is 31, the control value is set to 32 according to Table 2. The waveform codebook ROM 105 selects a group of output vectors corresponding to the group number 2 from the address counter 104, and outputs the output vector in the group to the distortion calculation circuit 106 every time the count value is sequentially incremented. Go. The count value from the address counter 104 is counted up 72 times, but since the control value is set to 32, the output vector is not output 32 times or more. When the count value is counted up to the number of control values, a reset signal is sent to the minimum distortion detection circuit 107 and the index latch 108. The distortion calculation circuit 106 obtains the distortion d according to the equation (3) using the minimum value separated normalization vector X and the output vector Y = {Y ₁ , Y ₂ , ..., Y ₉ } read from the waveform codebook ROM 105.

The distortion d obtained as described above is sent to the minimum distortion detection circuit 107. The minimum distortion detection circuit 107 reads the distortion d when the distortion d is smaller than before, reads the distortion d, sends the minimum distortion recognition signal to the index latch 108, and does nothing else. When the reset signal from the waveform codebook ROM 105 is input, the distortion d is updated.

The index latch 108 reads the count value from the address counter 104 when the minimum distortion recognition signal is sent, and when the reset signal from the waveform codebook ROM 105 is input, sends the read count value as an index value to the encoding circuit 109. .

The third table is a table relating to the grouping of coding in the coding circuit 109 which is judged by the maximum value from the amplitude normalization circuit 102.

The encoding circuit 109 has the minimum value from the minimum value separation circuit 102 and the maximum value from the maximum value regulation circuit 103 from the amplitude normalization circuit 103.
Group number 2 is selected by 31. Each group divided by the group number has a variable length coding code suitable for each group, and the index value from the index latch 108 is subjected to the variable length coding judged from the group number 2 and sent to the output signal terminal 110. .

In this embodiment, it is assumed that a 9-dimensional vector is input, the count value is set as shown in Table 1, the control value is set as shown in Table 2, and the absolute value of the difference is calculated in the distortion calculation. Variable-length coding is used for coding by using a value and setting a group number as shown in Table 3, but it is not limited to this.

As described above, according to the present invention, the input vector created from the input signal is reduced by the minimum value and normalized by the maximum value to reduce the number of patterns of the input vector.
Even if the number of output vectors in ROM is reduced, it can adapt to the input signal.

Further, the optimum output vector in the waveform codebook ROM can be obtained at high speed by using the position information of the elements in the input vector given the separated minimum value and maximum value. Furthermore, by limiting the number of output vectors in the waveform codebook ROM according to the size of the maximum value, it is possible to adapt to the input signal while improving the coding efficiency.

Since the means for performing the adaptation is performed only with the minimum value and the maximum value as described above, an extra amount of information is not required and therefore the coding efficiency is not reduced.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a vector quantizer according to an embodiment of the present invention, and FIG. 2 is a block diagram showing the configuration of a conventional vector quantizer. 102 ... Minimum value separation circuit, 103 ... Amplitude normalization circuit, 104
...... Address counter, 105 …… Waveform codebook ROM,
106 …… Distortion calculation circuit, 107 …… Minimum distortion detection circuit, 108 ……
Index latch, 109 ... Encoding circuit.

Claims (6)

[Claims] 1. A minimum value of the elements of the input vector is calculated by using an input vector that is a blocked input signal for each of k (k is an integer) samples in advance, and the minimum value is calculated from each element of the input vector. The minimum value separation circuit that outputs the minimum value separation vector that has been subjected to the minimum value separation process by subtracting it, and the maximum value of the elements of the minimum value separation vector are calculated, and from each element the An amplitude normalization circuit that outputs a maximum value and a minimum value separation normalization vector that is normalized by dividing the maximum value as an amplitude coefficient, and the minimum value separation normalization from the amplitude normalization circuit An address counter that sequentially outputs a count value that counts up to a predetermined number each time a vector is read out, and a vector that is normalized by separating the minimum value Storing previously optimized plurality of output vector based on statistical properties, sequentially reads out output vector each time the count value is counted up,
A waveform codebook ROM that outputs a reset signal when the count value is counted up by a predetermined number,
The minimum separation normalization vector and the waveform codebook
A distortion calculation circuit for calculating distortion with an output vector read from a ROM and outputting it as a distortion value, and the waveform codebook
When the reset signal from ROM is input, the distortion value is updated,
A minimum distortion detection circuit that outputs a minimum distortion recognition signal by sequentially comparing distortion values from the distortion calculation circuit and giving a minimum distortion, and a count value from the address counter by inputting the minimum distortion recognition signal. An index latch that reads and outputs the value being read as an index value by inputting a reset signal from the waveform codebook ROM, a minimum value from the minimum value separation circuit, and a maximum value from the amplitude normalization circuit A vector quantizer, comprising: an encoding circuit that performs predetermined encoding of the index value from the index latch. 2. The minimum value separation circuit is k in number (k is an integer) in advance.
, The input signal is a blocked input signal for each sample, the minimum value of the elements of the input vector is calculated, the position of the element of the input vector given the minimum value is output as the minimum value index, The minimum value separation vector subjected to the minimum value separation processing is output by subtracting the minimum value from each element and the minimum value is output, and the address counter reads the minimum value separation normalized vector from the amplitude normalization circuit. For each, the group number determined by the minimum value index and a count value that counts up to a predetermined number determined by the group number are sequentially output, and the waveform codebook ROM is the minimum value separation normalized vector statistics. Group that outputs a plurality of output vectors optimized in advance based on the physical properties of the address counter. Grouped according to the group number stored in the address counter, the group is selected corresponding to the group number from the address counter, and the output vector is sequentially read every time the count value is incremented, The vector quantizer according to claim 1, wherein the vector quantizer is configured to output a reset signal when the count value is counted up by a predetermined number corresponding to the number. 3. An amplitude normalization circuit calculates the maximum value of the elements of the minimum value separation vector output from the minimum value separation circuit, and outputs the position of the element of the vector given the maximum value as the maximum value index. Then, the maximum value is output together with the minimum value separation normalization vector that is normalized by dividing the maximum value as an amplitude coefficient from each element, and the address counter outputs from the amplitude width normalization circuit. Each time the minimum value separation normalization vector is sequentially read, the minimum value index output from the minimum value separation circuit and the group number determined by the maximum value index and the count value that counts up to the number determined by the group number Waveform codebook RO
M, a plurality of output vectors that have been optimized in advance based on the statistical properties of the minimum-value-separated and normalized vectors are grouped and stored according to the group number output by the address counter, and stored. When the group corresponding to the group number from the address counter is selected and the output value is sequentially read every time the count value is counted up, the count value is counted up by a predetermined number corresponding to the group number. The vector quantizer according to claim 2, wherein the vector quantizer is configured to output a reset signal. 4. The waveform codebook ROM fetches the maximum value from the amplitude normalization circuit even if the count value output from the address counter is counted up, and exceeds a predetermined control value judged by the maximum value. In this case, the reset signal is output without reading the output vector, and the encoding circuit performs predetermined encoding of the minimum value from the minimum value separation circuit and the maximum value from the amplitude normalization circuit, and the index latch 2. The index value from 1 is grouped according to the maximum value, and predetermined coding is performed based on the grouping. Vector quantizer. 5. The distortion calculation circuit calculates the absolute value of the difference between the respective elements of the minimum-value-separated and normalized input vector and the output vector from the waveform codebook ROM, and sets the sum of the absolute values as the distortion value. The vector quantizer according to claim 1, which outputs the vector quantizer. 6. The vector quantization according to claim 1, wherein the minimum value from the minimum value separation circuit, the maximum value from the amplitude normalization circuit and the index value from the index latch are variable length coded. vessel.