JPS58218242A - Mean value split vector quantizer - Google Patents

Abstract

PURPOSE:To perform encoding with high efficiency at a high speed, by dividing a signel space according to the mean value of a block previously and quantizing them by vectors. CONSTITUTION:An input signal sequence is divided into K (K: integer greater than 2) blocks and inputted to a register 5 as input vectors. At this point of time, a mean value computing element 6 calculates the amplitude mean value of the K blocks and sends the mean value to a code table switch 7. The switch 7 sets a threshold value to the mean value to decide which of output vector code table memories 9 and 10 is referred to; and output vectors are read out of the selected memory by a code table address counter 8 and compared with the input vectors, and element distortion is calculted by a parallel subtracter 12 and a parallel absolute value computing element 13. The maximum value of the element distortion is detected by a detector 14 while the minimum distortion is detected by a detector 15, and they are latched in an index latch 18 as the vector quantization outputs of the input vectors and inputted to a decoder. The decoder latches a signal from an encoder in the latch 18 to select the memory 9 or 10, obtaining an output vector 20.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a highly efficient vector quantizer that provides a quantization characteristic that provides minimum distortion based on the amplitude probability density of an input signal.

Conventional quantizers of this type employ scalar quantization to convert an input signal into a corresponding output signal level for each sandal.

FIG. 1 shows a conventional scalar quantizer. In the figure, (1) is the sequentially input signal sequence Machi, Nio,..., gx (K is an integer), +21 is the scalar quantizer, and (3) is the output signal sequence 11t, G,..., ! [is.

Assume now that the input signal sequence has a GaulS distribution in which the amplitude probability density of ) is centered at the origin. in this case.

The input signal sequence (1) is converted into the output signal sequence (2) for each sample.
) is quantized. In the scalar quantization characteristic that provides the minimum distortion, as shown in FIG. 2, the quantization level becomes coarser as the distance from the origin increases. However, if there is a correlation between each sample of the input signal sequence (1), the output signal sequence will not be optimally quantized even if it is quantized to the minimum distortion for each sample as in the past. .

This invention was made in order to eliminate the quantization loss caused by conventional scalar quantization, and the input signal sequence is grouped into blocks for each predetermined sample, and the blocks are collectively converted into blocks of the output signal sequence at high speed. With the purpose of providing a simple and highly efficient vector quantizer for conversion, we will now explain the principle of vector quantization. A process consisting of samples of the input signal sequence:
Input the bow vector X = '('F1. g2. "'
21).

The representative point of the dimensional signal space t including all input vectors
v2. −, uN).

1 each. The divisions of RK with RK as a representative point (for example, the center of gravity) are R1, R2, and R2, respectively. ...If RN, vector quantization Q
is defined by the following equation.

Q, RK→Y where R,i −Q−′(Yt) = (XBRK:
Q(X) = 111)U R1=R' + R
s FL) −〇 (' ~)) ml The above vector quantization Q can be regarded as a cascade connection of encoding C and decoding. At this time, the encoding C is R″(oy
is mapped to index set J-(1, 2, . . . , N), and decoding is mapping from J to Y. Namely.

0:R″-+J D:J-+Y Q=D-0.The index set, which is the encoded output, may be transmitted or recorded.

The above vector quantization is performed on JjJ,,i? , highly efficient encoding can be achieved.

Vector quantization, which converts the input vector λk into the output vector v1 with the minimum distortion, is performed by dividing the input vector I need to find it. This can be obtained by clustering using a model of the target image.

Vector quantization can be said to be the search for the output vector Vi that is closest to the input vector X (resulting in minimum distortion) from the set Y of output vectors. In this case, if the number of output vectors is large.

Vector quantization slows down the conversion speed. This drawback can be solved by dividing R1 in advance by the average value of the blocks and vector quantizing each. That is, it is a mixed form with scalar quantization. FIG. 3 shows an array of average divided output vectors in a two-dimensional signal space.

A block diagram of an embodiment of the mean value division vector quantizer according to the present invention is shown in FIGS. 4 and 5. Here, FIG. 4 shows an encoder, and FIG. 5 shows a decoder.

In the figure, (4) is an input vector, (5) is an input vector register, +6) is an average value calculator, (71 is a code table switcher, and (8) is a code table address counter.

+91 is the first output vector code table memory.

QG is the second output vector code table memory.

times is the output vector register, (1m is the parallel subtractor, 11
3 is a parallel absolute value calculator, α( is a maximum element distortion detector, 0
51 is a minimum distortion output vector detector, ue is an index signal, (171 is a code table switching signal, 2 is an index latch, ill is an encoder output signal, and ■ is an output vector.

First, the operation of the encoder shown in FIG. 3 will be explained.

The input signal sequence of the encoder is grouped into blocks and input to the encoder X = (Xs, Z2..., ff
1K) into the input vector register (5). At this point, there are 2 average value calculators (6) (K is 2
(integer greater than or equal to)) and sends the average value to the code table switch (7). The code table switch 17+ sets a threshold on this average value and determines which output vector code table memory to refer to. Next, the code table address counter (8) sequentially inputs the output vector vi from either the selected first output vector code table memory (91 or the second output vector code table memory Q11). = 1.2....,N
Read up to This output vector fz is the input vector
is compared with the element type Did m j Vi X for each element.
il j (here 1=1.2.·, K) is calculated through a parallel subtracter α2 and a parallel absolute value calculator αj. Next, the maximum value of the element type Dig is detected by the maximum element distortion detector [+41, and the maximum element type D of the output vector V% is
Let it be i. The maximum element type is determined for each of the output vectors Ili that are read out sequentially, and the minimum value of each maximum element type is determined as the minimum distortion by the minimum distortion output vector detector. That is, the minimum distortion is D = Min (Mas
j Vi XiJ j 〕l. This output vector resulting in the minimum distortion is the vector quantization output of the input vector X. When the minimum distortion output vector is detected, the minimum distortion output vector detector u9 sends a strobe signal to the index latch, and outputs the code table switching signal αη and the index signal (1e is the code table address of the output vector) to the index latch Q.
be taken in by l.

The index signal ae and the code table switching signal a
? l is the encoder output signal (transmitted or recorded as 11)
It is reproduced and becomes the decoder input.

next! The operation of the decoder shown in Fig. J5 will be explained.

The encoder power signal sent from the encoder is input to the index latch QIS of the decoder. For the encoded output signal US, the index signal αB becomes the address signal of the first output vector code table memory OI or the second output vector code table memory OI, and the code table switching signal aη refers to which output vector code table memory. This is selection information on whether to do so or not. The output vector Ili selected and read out from the address is latched in the output vector register Uυ, and the output vector Ili is obtained as a decoded output.

As described above, the average value dividing vector quantizer according to the present invention divides the signal space in advance according to the average value of the block and refers to the output vector code table.
Even if the output vector doubles, the reference time does not increase.

Coding efficiency η is calculated by multiplying the number of samples in the block by 9. The number of output vectors in one output vector code table N = 2
', the number of output vector code tables (corresponding to the number of average threshold levels) is assumed to be T=2.

s+t 7=　□　Bit/Samp〃 to It is.

Note that in cases of 2 or more, the code table of the output vector is switched based on the average value of the block, but a medium value may be used instead of the average value. Furthermore, although minimax approximation is used to calculate distortion between input and output vectors, it is of course possible to use Euclidean norm (squared error) or absolute value norm.

Furthermore, although the present invention aims to increase speed by parallel processing of vector operations, if the data rate of the input signal is slow,
Of course, sequential processing may be performed using a microprocessor or the like.

Further, in the present invention, two output code tables are switched, but the output code table may be divided into a larger number of tables.

As described above, in the average value dividing vector quantizer according to the present invention, the signal space is divided in advance based on the average value within a block, and by vector quantizing the signal space, high-speed and high-performance encoding can be achieved. There are effects that can be achieved.

[Brief explanation of drawings]

Fig. 1 is an explanatory diagram of a conventional scalar quantizer, Fig. 2 is an explanatory diagram of quantization characteristics of conventional scalar quantization, and Fig. 3 is a two-dimensional signal of the mean value division vector quantizer according to the present invention. An explanatory diagram showing the arrangement of output vectors in space, FIG. 4 is a block diagram showing an embodiment of the encoder of the mean value division vector quantizer according to the present invention, and FIG. 5 shows the mean value division according to the present invention. FIG. 2 is an explanatory diagram showing an example of a decoder of a vector quantizer. In the figure, (1) is the input signal sequence, (21 is the scalar quantizer, (3I is the output signal sequence, (4I is the input vector, +
5) is an input vector register, L6) is an average value calculator,
L7) is a code table switch, (8) is a code table address counter, L9) is the first output vector code table memory, +II is the second output vector code table memory, tlB is the output vector register,
ag is a parallel subtractor, (13 is a parallel absolute value calculator, 0
4 is the maximum element distortion detector, tap is the minimum distortion output vector detector, ae is the index signal, (171 is the code-to-chip switching signal, OlO is the index latch, 9 is the encoder output signal, ■ is the output It is a vector.In addition, the same or corresponding parts in the two figures are indicated with the same reference numerals.Agent Shin Kuzuno - !11th officer Figure 1 Figure 3 χ2゛

Claims (1)

[Claims] 1. After dividing the input vector into blocks of a predetermined number of samples (K is an integer of 2 or more) of the input signal, the 1-dimensional signal space is scalar-divided into a plurality of samples using the average value of the blocks. a plurality of output vector code table memories that determine and store a plurality of sets of output vectors that produce minimum distortion based on the distribution of input vectors; an average value calculator that calculates the average value of a block of input signals; an output vector code table memory switch that selects one of the plurality of output vector code memories based on the average value of the block; and a code table address counter that sequentially reads output vectors from the selected output vector code table memory; A min-max operator that compares the input vector with the output vector read out sequentially and detects the output vector with the minimum distortion by min-max approximation, and the output vector code table address of the output vector with the minimum distortion and the selected output. The present invention is characterized by comprising an encoder that encodes a vector code table memory number, and a decoder that reads output vectors corresponding to input vectors from a plurality of output vector code table memories based on the encoder output signal. A mean-split vector quantizer.