JPH03212084A - Band split type vector quantization system - Google Patents

Abstract

PURPOSE:To reduce quantization loss by applying quantization to picture elements at the same location specially in a signal series at a different band in a lump. CONSTITUTION:A band split means is realized by 3 band split filters 2, 5, 6 and a picture signal series 1 is split into 4 series. That is, the picture signal series 1 is split into series included in two bands by the 2nd band split filter 5 and the 3rd band split filter 6 and split in total 4 series. Then a vector quantization coder 14 receives a 2nd low frequency series 7, a 2nd high frequency series 8, a 3rd low frequency series 9 and a 3rd high frequency series 10 to apply vector quantization to picture elements in a lump at the same location specially in each series and outputs a vector index 15. Thus, the quantity of information is reduced without decreasing quantization level number.

Description

[Detailed Description of the Invention] [Industrial Application Field] This invention is a band division method in which an image signal sequence is divided into a plurality of sequences for each band, and pixels at the same spatial position in each sequence are vector quantized. It is related to type vector quantization method.

[Conventional technology]

FIG. 6 shows, for example, sub-band coding monochrome and color images (Sub-Band Codin
g of Monochrome and C
o1or Images)+ Aiiiiiii
IEEE Transactions on Cir.
35, No. 2 (1988), in which the encoding method is shown in blocks. In the figure, ■ is a digitized image signal sequence, and 2 is a quadrature mirror filter (QMF
; Quadrature Mirror Filter
r), 3 is the first low-pass sequence divided by the first band-splitting filter 2, 4 is the first high-pass sequence, and 5 is the first low-pass sequence 3. A second band division filter further divides by QMF, 7 is the second low frequency sequence divided by the second band division filter 5, 8 is the second high frequency sequence, and 6 is the first high frequency sequence. 4 is further divided by QMF, 9 is the third low frequency sequence divided by the third band division filter 6, 10 is also the third high frequency sequence, 1
1 is a prediction signal sequence that predicts the second low-frequency sequence 7, I2 is a subtracter that calculates a prediction error, 13 is a prediction error signal sequence,
30 is a first quantizer that quantizes the prediction error signal sequence 13; 18 is an adder that adds and decodes the prediction signal sequence II and the first quantized output 34; and 19 is a decoded second quantizer. 20 is a predictor that forms the predicted signal sequence 11 from the decoded sequence 19, 31 is a second quantizer that quantizes the second high-frequency sequence 8, and 32 is a third quantizer. A third quantizer that quantizes the low frequency sequence 9, 33 a fourth quantizer that quantizes the third high frequency sequence IO, 34 a first quantization output, and 35 a second quantizer. 36 is the third quantization output, 37 is the fourth quantization output, 38 is the first quantization output 34, the second quantization output 35, the third quantization output 36 and the fourth quantization output. A multiplexer that multiplexes the quantized output 37, 39 a multiplexed sequence, 40 a first one from the multiplexed sequence 39;
a demultiplexer that separates the quantized output 34, the second quantized output 35, the third quantized output 36, and the fourth quantized output 37; and the second quantized output, i.e. the quantized second
a second band synthesis filter that synthesizes the high frequency sequence 35 of
25 is a third band synthesis filter that combines the third quantized output, that is, the quantized third low-frequency sequence 36, and the fourth quantized output, that is, the quantized third high-frequency sequence 37; 26 is the first filter obtained from the second band synthesis filter 24.
, 27 is the first decoded high frequency sequence obtained from the third band synthesis filter 25, and 28 is the first decoded high frequency sequence that combines the first decoded low frequency sequence 26 and the first decoded high frequency sequence 27. 1 is a band synthesis filter, and 29 is a decoded image signal sequence.

Next, the operation will be explained. The image signal sequence l is divided into sequences included in two bands by the first band division filter 2, the second band division filter 5, and the third band division filter 6, for a total of four bands.

Since each band divided into two by QMF has a bandwidth of 1/2, 2:1 subsampling is performed. Now, the first
Assuming that the band division filter 2 is applied in the vertical direction of the image, and the second band division filter 5 and the third band division filter 6 are applied in the horizontal direction of the image, these two band division filters apply the two-dimensional QMF. Configure. At this time, if the original image has MXN pixels, the second low frequency range...Horizontal direction = low frequency range/Vertical direction = low frequency range.

M/2 x N/2 pixels Second high range...Horizontal direction = high range / Vertical direction = low range.

M/2 x N/2 pixels Third low range...Horizontal direction - low range/vertical direction = high range.

M/2 XN/2 pixels Third high range...Horizontal direction = high range / Vertical direction - high range.

M/2 XN/2 pixels The image is divided into small images included in each of the four bands.

Then, each divided image is individually encoded.

Since the second low band is the lowest of the four bands, the correlation is strong, and it is effective to apply predictive coding to images within this band. s for the second low frequency series 7;
It shall be expressed as (t=1+2+...). Here, i
is a number representing the order of the signal sequence. Furthermore, if the prediction signal sequence 11 is expressed as P = (i = 1.2+...), the prediction error signal sequence 13 is s, -p.

It is expressed as This will be expressed as ε=(1""L2+...). The first quantizer 30 converts the prediction error signal sequence ε, 13 into a differential PCM (DPCM) of a predetermined number of levels.
) code is used to quantize. An example of quantization characteristics is shown in FIG. The first quantization output 34, which is a quantized DPCM signal sequence, contains quantization errors. The first quantized output 34, which is a DPCM signal sequence? x < i = 1
121...), and the quantization error is expressed as qN, ε, = s, -p.

? ,=ε=+(lx It is.

The adder 18 adds the predicted signal sequence P, 11 and the first quantized output t, 34 to obtain a decoded sequence 19. Assuming that the decoded sequence 19 is Sdi (i = 1121...), the decoding operation is expressed as follows.

5at=P=+ft=34+qN The decoding operation of predictive coding is exactly the same on both the transmitting side and the receiving side.

Since each signal sequence of the second high frequency series 8, the third low frequency series 9, and the third high frequency series 10 includes image changes (high frequency components), they are directly PCM quantized. , a second quantizer 31, a third quantizer 32 and a fourth quantizer 33 are used.

The multiplexer 38 sends out a multiplexed sequence 39 in which the first quantized output 34, the second quantized output 35, the third quantized output 36, and the fourth quantized output 37 are multiplexed.

On the receiving side, the demultiplexer 40 demultiplexes the multiplexed sequence 39
A first quantized output 34, a second quantized output 35, a third quantized output 36, and a fourth quantized output 37 are separated from the quantized output. The decoding sequence 19 of the lowest band is obtained by the same decoding operation as on the transmitting side.

The second band synthesis filter 24 synthesizes the decoded sequence 19 and the second quantized output 35, and the third band synthesis filter 25
combines the third quantized output 36 and the fourth quantized output 37. Each combined sequence is combined in a first band synthesis filter 28 to obtain a decoded image signal sequence 29. Furthermore, 1:2 interpolation is performed every time bands are combined.

When the quantization characteristics shown in FIG. 7 are used in the first quantizer 30, the second quantizer 31, the third quantizer 32, and the fourth quantizer 33, an image of MXN pixels The amount of information required to encode is 10gz 16 x 4 x (M/2x N/2)
= 4 MN (bits).

[Problem to be solved by the invention]

Conventional band division encoding methods are configured as described above, and since pixels in multiple bands are individually encoded, it is necessary to reduce the number of quantization levels in order to reduce the amount of information after encoding. As a result, there was a problem in that image quality deteriorated.

This invention was made in order to solve the above-mentioned problems, and to obtain a band-splitting vector quantization method that can reduce the amount of information after encoding without drastically lowering the number of quantization levels. With the goal.

[Means to solve the problem]

The band division vector quantization method according to the present invention divides an input image signal sequence into n (n is an integer of 2 or more) frequency band components by a band division means, and then divides the input image signal sequence into n (n is an integer of 2 or more) frequency band components. Among the pixels in the image, n pixels located at the same spatial position are collectively vector quantized by a vector quantizer.

(Function) The vector quantizer of the present invention reduces quantization loss by collectively quantizing pixels located at the same spatial position in signal sequences in different bands.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings. 1st
In the figure, 14 inputs a second low-frequency series 7, a second high-frequency series 8, a third low-frequency series 9, and a third high-frequency series 10, and inputs the spatially identical A vector quantization encoder (vector quantizer) that performs vector quantization on pixels at a position and outputs a vector index 15, 16 inputs the vector index 15 indicating the vector quantization result, and outputs the vector index 15
50 is a quantized second low frequency sequence, 52 is a quantized second high frequency sequence, and 53 is a quantized third high frequency sequence. , 54 is the quantized third high-frequency series, and the other parts are the same as those shown in FIG. 6 with the same reference numerals. In this embodiment, the band division means is realized by three band division filters 2, 5.6,
The image signal sequence l is divided into four sequences.

Next, the operation will be explained. The image signal sequence 1 is divided into sequences included in two bands by the first band division filter 2, the second band division filter 5, and the third band division filter 6, for a total of four sequences.

Since each band divided into two by QMF has a bandwidth of 1/2, 2:1 subsampling is performed. It is assumed that the first band division filter 2 is applied in the vertical direction of the image, and the second band division filter 5 and the third band division filter 6 are applied in the horizontal direction of the image. At this time, the original image (MXN
pixel), the second low frequency...Horizontal direction = low frequency / Vertical direction = low frequency M/
2 XN/2 pixels Second high range...Horizontal direction = high range/Vertical direction - low range M/
2 x N/2 pixels 3rd low range...Horizontal direction - low range/vertical direction = high range M/
2 x N/2 pixels Third high range...Horizontal direction = high range / Vertical direction = high range.

M/2 XN/2 pixels Division into each small image included in the four bands is performed.

Then, the vector quantization encoder 14 encodes each sequence 7.8
, 9.10, vector quantization is performed by grouping the pixels at the same spatial position, and the vector index is 15.
Output.

Further, since the second low band is the lowest band among the four bands, the correlation is strong, and it is effective to apply predictive coding. Now, as shown in FIG. 2, it is assumed that a certain pixel in the second low frequency band is to be encoded. At this time, consider pixels that are spatially located at the same position in the second high frequency band, the third low frequency band, and the third high frequency band. Hereinafter, for the sake of explanation, the second low range is LL, the second high range is HL, and the third low range is LF (, third
The high range of is expressed as HH. L to be encoded
If a pixel on L is a flat part of the image, the prediction error in LL is small. At the same time, since pixels belonging to HL and LHHH are in the high range, they take small values in flat areas. On the other hand, if the pixel to be encoded is a changing part of the image,
The prediction error in LL becomes large. At the same time, HL, L
Pixels belonging to H and HH take some value that is not small. In this way, there is a correlation between changes in pixels in the low-frequency bone and pixels in the high-frequency region. Therefore, if vector quantization is applied after predictive coding is performed on the second low-frequency sequence 7, the coding efficiency is further improved.

FIG. 3 is a block diagram showing a band division type vector quantization method constructed based on this idea. Here, the second low-frequency series 7 corresponding to LL is defined as LL=(
1=1+2+...). i is a number representing the order of the signal sequence. Furthermore, the predicted signal sequence 1
1 is expressed as Pi (i = 1.2...), the prediction error signal sequence 13 is expressed as LL, -, P. This is ε
= (i=t+ 21...). In addition, the second high frequency series 8 corresponding to HL is HL8, L
The third low frequency series 9 corresponding to H is expressed as LH, and the third high frequency series lO corresponding to HH is expressed as HH. same i
It is assumed that pixels with 1 and 2 correspond to the same position in space.

The vector quantization encoder 14 has ε,,HL□.

LH,, HH, are collectively treated as a vector, the closest one is determined from among the output vectors, and the vector index 15 of the output vector is output. An example of the output vector group is shown in FIG.

Vector quantization decoder 16 outputs an output vector corresponding to vector index 15.

The output vector consists of a quantized prediction error sequence 51, a quantized second high frequency sequence 52, a quantized third low frequency sequence 53, and a quantized third high frequency sequence 54. The adder 18 adds the predicted signal sequences P, 11 and the quantized prediction error sequence 51 to obtain a decoded sequence I9. The above decoding operation is exactly the same on both the sending and receiving sides.

On the receiving side, the second band synthesis filter 24 synthesizes the decoded sequence 19 and the quantized second high frequency sequence 52, and the third band synthesis filter 25 synthesizes the quantized third low frequency sequence 52. The sequence 53 and the quantized third high frequency sequence 54 are combined. Each synthesized sequence is passed through the first band synthesis filter 2
8, and a decoded image signal sequence 29 is obtained. A 1:2 interpolation is performed each time bands are combined.

Then, when using the quantization characteristics shown in Fig. 4, M
The amount of information required to encode an image of XN pixels is 1ogz 16 X (M/2x N/2) = M-
N (bits).

In the above embodiment, the band is divided into four, but the number does not need to be four if the band is divided into a plurality of bands.

Furthermore, in the above embodiment, the case where the low-frequency prediction error series 51 and the other series 52, 53, and 54 are vector quantized together is explained, but for example, the prediction error series 51
Only the other sequences 52, 53, and 54 may be vector quantized together. A configuration diagram in this case is shown in FIG.
is input and the vector index 15 is output. The multiplexer 55 then outputs the vector index 15
and the first quantized output 34 to produce a multiplexed sequence 56.
Output.

On the receiving side, a multiplexer 57 separates the multiplexed sequence 56 into a vector index 15 and a first quantized output 34 . Then, the vector quantization decoder 16 outputs a second high frequency sequence 52, a third low frequency sequence 53, and a third high frequency sequence 54, and the adder 18 outputs a decoded sequence 19.

In addition, in the method shown in FIG. 3 or FIG. 5, predictive coding is applied only to the second low-frequency sequence 7, but it is also possible to apply predictive coding to the other sequences 8, 9, and 10. good.

〔Effect of the invention〕

As described above, according to the present invention, the band division type vector quantization method 2 divides an image signal sequence into a plurality of sequences for each band by the band division means, and divides the pixels included in these sequences into Since the configuration is such that pixels located at the same spatial location are collectively vector quantized using a vector quantizer,
This has the effect of reducing the amount of information without reducing the number of quantization levels.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing a band-splitting vector quantization method according to an embodiment of the present invention, FIG. 2 is an explanatory diagram for explaining the encoding operation, and FIG. 3 is a diagram according to another embodiment of the present invention. FIG. 4 is a characteristic diagram showing an example of vector quantization characteristics; FIG. 5 is a configuration diagram showing a band-splitting vector quantization method according to still another embodiment of the present invention. 6 is a block diagram showing a conventional band division type encoding system, and FIG. 7 is a characteristic diagram showing an example of the characteristics of the quantizer shown in FIG. 6. 2 is a first band division filter, 5 is a second band division filter, 6 is a third band division filter, 14 is a vector quantization encoder, 16 is a vector quantization decoder, 20 is a predictor, 24 is a second band synthesis filter, 25 is a third band synthesis filter, 28 is a first band synthesis filter, 3
0 is a first quantizer, 55 is a multiplexer, and 57 is a demultiplexer. In addition, in the figures, the same reference numerals indicate the same or equivalent parts.

Claims (1)

[Claims] band dividing means for dividing an image signal sequence into a plurality of bands; and vector quantization for vector quantizing blocks of pixels that are spatially located at the same position among the pixels included in the sequence for each of the plurality of bands. Band-splitting vector quantization method with