JPS5994936A - Vector quantizing method - Google Patents

Abstract

PURPOSE:To attain the reduction in average distortion by scaler quantization with the same information amount by executing a two-stage quantization method using a specific vector dictionary. CONSTITUTION:The vector quantizing section 21 decides to which an input vector belongs. When the input vector S belongs to a group 25a of a center of gravity point C1, a dispersion vector of the group 25a is represented by a barycentric vector representing the C1, and a coordinate with minimized distortion is selected from a lattice point 26 independent of each coordinate expressed by (B-B0) formed by using the dispersion vector. In this case, B-B0=4 and 16 lattice points 26 in total by 2-bit per coordinates x1, x2 are provided around the center of gravity C1. The number of the group 25 is transmitted by using B0 bits and the number of the lattice point 26 is transmitted from terminals 27, 28 to the decoder 13 as an output of the coder 11 by using (B-B0) bits. Thus, the two-stage quantization using 2B<0>X2 sets (B0<B, where B is the quantized bit number) of vector dictionaries is executed, the average distortion is made less than the scale quantization by using the same amount of information.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a vector quantization method suitable for cases where the amount of quantization information per vector is large.

<Prior art> In general, vector quantization converts each sample value into a vector Ex"("11 e12-...etp)
, E2=(e+1e211 =-e!Ip) 0°1
°°−En = (ent ensoo“0′.

enp), and the representative vector F1=(ft□f1□・・・flp
), F2"(f'1fzg-"・fgp)
”...Fm=('mt'mz...f
The one with the minimum distortion, that is, the closest one, is selected from the dictionary of mp), and that number is used as the code. In this way, since one vector is encoded using one representative value, such as a number, efficient encoding is possible when the number of one vector element is small, and especially the encoding per one vector is possible. If the amount of information in is B (bit), then
If a dictionary consisting of 2B representative vectors is created to reflect the statistical distribution of data from learning samples, it is possible to efficiently encode data that has an unusual or biased distribution. The average distortion due to encoding can be reduced in proportion to 2-(2B/P)K, where P is the vector dimension.

However, for a given amount of information B, a dictionary of 2B vectors and more than 10 times as many learning samples are required.
is realistically 10 or less.

For example, if B≧20, it can be said that it is impossible to create a dictionary even with any existing electronic computer. That is, in general vector quantization, there is a limit to reducing the average distortion associated with encoding due to the restriction on the amount of information B per l vector.

Furthermore, when B is different, it is necessary to create a separate dictionary for each value of B.

This invention takes advantage of the advantage of vector quantization, which allows efficient quantization of data even if the distribution is biased,
Moreover, the present invention provides a vector quantization method that can reduce distortion accordingly even when the encoded information amount B is 20 or more.

<Embodiment> FIG. 1 shows an example of an information transmission system to which the vector quantization method of the present invention is applied. The encoder 11 is connected to a decoder 13 via a communication path 12, and dictionaries 14 and 15 are connected to the encoder 11 and the decoder 13, respectively. The dictionary 14 includes a centroid vector section 16 in which centroid vectors representing representative values are stored, and a dispersion vector section 17 in which dispersion vectors indicating the distribution of villages with respect to each centroid are stored. The dictionary 15 on the decoder side has exactly the same configuration as the dictionary 14 on the encoder side, and includes a centroid vector section 18 and a dispersion vector section 19. The encoder 11 includes a vector quantizer 21 and a scalar quantizer 22, each using a centroid vector section 16 and a variance vector section 17.

The vector quantization method of the present invention will be explained using two-dimensional (n=2) data having the distribution shown in FIG. 2 as an example. Second
Assume that data is distributed within a region surrounded by a closed curve 23 in the figure. Normal XL.

As long as X2 coordinate independent quantization or quantization after coordinate transformation is scalar quantization, as long as quantization is performed for each data,
Distortion increases for a given amount of information.

However, in this invention, vector quantization is performed in two stages. Input data in which a sample value series is made into a vector whose elements are the sample values for each block is input to the vector quantization unit 21 from the input terminal 24, and is vector quantized with Bo=2. That is, the entire area is divided into 20 areas 25a to 25d in accordance with the a priori distribution of the area 23, and the center of gravity C1 to C4 of each block, that is, each village 25a to 25d.
has been created as a dictionary. In the vector quantization unit 21, the input vector is 2B (one village 258 to 2
It is determined whether it belongs to 5d. For example, it is assumed that the input vector S belongs to the village 25a at the center of gravity C0.

The dispersion vector of the village 25a is represented by a centroid vector indicating its centroid C1. The one with the minimum distortion is selected from the coordinate-independent grid points 26 expressed by (B-Be) pits created using this dispersion vector. In the example in the figure, B-B = 4 and 2 bits per each coordinate x1°x2, totaling 1
Six lattice points 26 are provided around the center of gravity C1. B
The 0 bit is used to convey the number of the village 25, and the (BBo) pit is used to transmit the lattice point 260 number as the output of the encoder 11 to the decoder 13 from the terminal 27°28.

In the decoder 13, B in the sent B-bit code.

The center of gravity vector part 18 and the distributed vector part 19 are referred to using the pit village number and the grid point number of the B-B0 pit, respectively, and in this example, the vector of the grid point 26H near the input vector S is decoded and the terminal Output to 31.

During the first stage of vector quantization, multiple villages are selected as candidates, and for each candidate, the grid point with the minimum distortion is selected, and finally the combination of the ones with the minimum distortion is sent. In other words, the average distortion can be further reduced with the same amount of information.

In addition, the dispersion vector section 17.19 has a value for each coordinate for each centroid point, but even if each is replaced with an average value, the overall distortion will not become very large, so a dictionary of dispersion vectors can be easily used as needed. It is possible to make it into something.

<Effects> As described above, according to the present invention, the two-stage quantization method using two 2BOX vector dictionaries allows the average distortion to be smaller than that of scalar quantization with the same amount of information. A comparative example in the case of audio sample values is shown in FIG. The horizontal axis shows the number of bits per sample B, the vertical axis shows the S/N ratio, and the curve 32
is for one-dimensional optimal quantization (conventional case), curve 33 is for six-dimensional vector and Gaussian quantization, and curve 34 is for one-dimensional optimal quantization (conventional case).
This is the case for two-dimensional vectors and force-force quantization. If the number of pits is larger than this, vector quantization becomes difficult, but with the present invention, vector quantization is possible and the S/N ratio is also good. The difference is even more significant for data with unique distributions. According to the quantization method of the present invention, it is expected that the distortion will be on the same level as the distortion of vector quantization using 2B vector dictionaries. Encoding information amount B
When B is 20 or more, a 2B dictionary is unrealistic, so the vector quantization method of the present invention is extremely effective for quantization when B is 20 or more.

Furthermore, when B changes, the vector quantization method of this invention eliminates the need to change the dictionary as long as the rules for setting grid points are determined, and the amount of information per sample value (B/P)
can be precisely controlled. In normal scalar quantization, the amount of information per sample value is basically limited to an integer value, and in normal vector quantization, a separate dictionary must be prepared for each change in B.

[Brief explanation of the drawing]

Figure 1 is a block diagram showing an example of an information transmission system that encodes and transmits input data to which the vector quantization method of the present invention is applied, and Figure 2 is an example of quantization of data with a two-dimensional biased distribution. FIG. 3 is a diagram showing an example of comparison with a scalar quantization method using sample values in the audio frequency domain. ll: encoder, 12: communication path, 13: decoder, 14.1
5: Dictionary, 16.18: Centroid vector section, 17, 19
: Dispersion vector section, 21: Vector quantization section, 22 NIS force quantization section. Patent applicant: Taku Kusano 205, agent of Nippon Telegraph and Telephone Public Corporation

Claims (1)

[Claims] (1) Part B of the number of quantization bits B allocated to each vector in a vector quantization method in which a sequence of sample values is quantized block by block. (B > Bo, BlBo is an integer) to select N candidate vectors from a dictionary consisting of 2 B 6 samples in descending order of distortion; The second quantization stage quantizes each candidate vector independently on the coordinate axes using (B-Be) bits based on the dictionary of variance vectors or the average variance value. A vector quantization method characterized by determining the vector with the minimum distortion.