JPH0642736B2 - Interframe vector quantizer - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an interframe vector quantizer for data compression of a moving image signal by vector quantizing an interframe difference signal.

[Conventional technology]

FIG. 4 is a block diagram showing a configuration example of a conventional inter-frame vector quantizer shown in, for example, “Murakami et al.,“ Vector Quantization Method Inter-frame Coding Simulation ”(Proceeding 1175, National Conference of the Institute of Electronics, Information and Communication Engineers). It is a figure.
In the figure, (1) is an input image signal sequence, (2) is a subtractor that subtracts from the inter-frame prediction signal, (3) is the inter-frame prediction signal, (4) is the inter-frame difference signal, and (39) is Vector quantization coding unit, (40) coded data, (41) vector quantization decoding unit, (16) decoded frame difference signal, (17) decoded frame difference signal (16) An adder for adding the inter-frame prediction signal (3), (18) a decoded image signal sequence,
(19) is a frame memory that gives a frame delay to the decoded image signal sequence (18) to form an inter-frame prediction signal (3),
(20) is a variable length coding unit, (21) is a buffer for speed smoothing, (22) is a threshold value, (23) is a line I / F (interface), and (24) is a transmission signal. is there.

Further, FIG. 5 is a block diagram showing a configuration example of the vector quantization coding unit (39). In the figure, (28) is a mean value separation normalization unit, (29) is a normalization vector, (31) is a codebook that stores the output vector, (32) is an output vector, and (33) is the normalization vector. Vector (29) and the output vector
(32) distortion calculating section for calculating distortion, (36) distortion calculated by the distortion calculating section, (37) minimum distortion detecting section for detecting the minimum distortion from the distortion (36), (30 ) Is the average value separation normalization unit (28)
Average value and amplitude separated by, (34) is the block discriminator,
(22) is the threshold used for block identification, (35) is the block identification information, (38) is the output vector index that gives the minimum distortion, and (40) is the encoded data.

Next, the operation will be described. The input image signal sequence (1) is
The inter-frame prediction signal (3) is subtracted by the subtractor (2) and converted into the inter-frame difference signal (4). Since the inter-frame difference signal has smaller power than the original signal, it is a signal that can be encoded with a small encoding error.
A vector quantization coding unit for the inter-frame difference signal (4)
It encodes in (39). At this time, the threshold value (22) is used as a parameter. Vector quantization decoding unit (39) to the encoded data (40) coded in the vector quantization coding unit (39)
In, the decoding is performed and the inter-decoded difference signal (16) is obtained.
An adder (17) adds the inter-frame prediction signal (3) and the decoded inter-frame difference signal (16) to obtain a decoded image signal sequence (18). The decoded image signal sequence (18) is stored in the temporary frame memory (19) and given a frame delay to form an inter-frame prediction signal. On the other hand, the encoded data (40)
Is subjected to variable length coding in the variable length coding unit (20), temporarily stored in the buffer (21), subjected to speed smoothing processing, and then transmitted through the line I / F (23) to the transmission signal (24). Is sent as. The buffer (21) outputs a threshold value (22) proportional to the amount of accumulated variable-length coded data and gives it to the vector quantization coding unit (39) to control the amount of information generated. . Coding and control of the amount of information generated in the vector quantization coding unit (39) will be described. The input signal to be vector-quantized is the inter-frame difference signal (4).
The signal (4) is block (vector) -ized in the average value separation / normalization unit (28) and subjected to the average value separation / normalization process. If the blocked input signal is represented by S = [S ₁ , S ₂ , ..., S _k ], the average value separation normalization process is described as follows, for example.

The normalized vector X = [x ₁ ,
x ₂ , ..., x _k ] is a vector S before the mean value separation normalization by separating the scalar values of the mean value and the amplitude.
Compared with, the probability distribution is made more uniform, which has the effect of improving the efficiency of vector quantization described below. Normalization vector (2
The distortion between the output vector (32) read out from the code book (31) and the code book (31) is obtained in the distortion calculator (33). Minimum strain detector (37)
Then, among the distortions (36) of the output vector and the input vector (29) stored in the code book (33), the smallest one is detected and the index number (38) of the output vector that gives the minimum distortion. Is output. This process is vector quantization. Written as an expression, it becomes as follows.

Distortion d: However, y _i = [y _i1 , y _i2 , ..., Y _ik ] is the output vector Y = [y ₁ , y ₂ , ..., Y _i , ... y _N ] is the codebook content vector quantization Q: Q (X) = Y _i At this time, the encoding process is a mapping from x to i, and the mapping from i to y _i (reading of code book) is the decoding process. i corresponds to the index (38). On the other hand, the average value and the amplitude (30) are used in the block identification circuit (34) together with the threshold value (22) for block identification. The block identification is represented as follows, where the threshold value (22) is Th.

| M | ≤Th and g≤Th ... invalid block | m |> Th or g> Th ... valid block For an invalid block, the inter-frame difference signal of the block is treated as 0. Therefore, at this time, the average value and the amplitude (3
0) and index (38) need not be transmitted. Coded data (40) output from the vector quantization coding unit (39)
Consists of mean value and amplitude (30), block identification information (35) and index (38), but in the case of an invalid block, only the block identification information (35) has a meaning, so the threshold ( 22) makes it possible to control the amount of information generated.

[Problems to be solved by the invention]

Since the conventional inter-frame vector quantizer is configured as above, the control range of the information generation amount is small, and if the information generation amount is intentionally suppressed to a minimum, the changed part on the screen becomes an invalid block. There was a problem that the so-called "sticky noise" remained.

The present invention has been made to solve the above problems, and an object of the present invention is to obtain an interframe vector quantizer capable of significantly controlling the amount of information generation without significantly impairing the quality of a reproduced image. To do.

[Means for solving problems]

In the inter-frame vector quantizer according to the present invention, the vector quantizer that performs the encoding process has a multi-stage configuration, and the elements that make up the vector in the vector quantization in the previous stage are the vectors in the vector quantization in the next stage. (Block) is the average value.

[Action]

The inter-frame vector quantizer of the multi-stage configuration according to the present invention can significantly reduce the information amount by using only the front stage of the multi stages for encoding and passing the latter stage when it is desired to suppress the information generation amount. When there is a margin in the amount of information generated, all stages are used for encoding, and the result of vector quantization in the preceding stage does not adversely affect the vector quantization in the succeeding stage.

Example of Invention

An embodiment of the present invention will be described below with reference to the drawings. First
FIG. 3 is a block diagram showing a configuration example of the inter-frame vector quantizer coding unit according to the present invention. In the figure,
(1) is the input image signal sequence, (2) is a subtractor that subtracts from the inter-frame prediction signal, (3) is the inter-frame prediction signal, (4) is the inter-frame difference signal, and (5) is the first-stage vector quantum. Encoding unit, (6) first stage encoded data, (7) first stage vector quantization decoding unit, (8) first stage decoded signal, (9) inter-frame difference signal, (4) and first stage A subtractor for obtaining the difference from the decoded signal (8), (10) is the first stage error signal, (11) is the second stage vector quantization coding unit, (12) is the second stage coded data, and (13) is Second stage vector quantization decoding unit, (14) is the second stage decoded signal, (15)
Is an adder for adding the first-stage decoded signal (8) and the second-stage decoded signal (14), (16) is a difference signal between decoded frames, (17)
Is an adder for adding the inter-frame prediction signal (3) and the inter-frame difference signal (16), (18) is a decoded image signal sequence,
(19) is a frame memory that gives a frame delay to the decoded image signal sequence (18) to form an inter-frame prediction signal (3),
(20) is a variable length coding unit, (21) is a buffer for speed smoothing, (22) is a threshold, (23) is a line I / F (interface), and (24) is a transmission signal. Is. FIG. 2 is an explanatory diagram for explaining the relationship between the first-stage vector quantization and the second-stage vector quantization. FIG. 3 is a block diagram showing a configuration example of the first-stage vector quantization coding unit (5) in FIG. In the figure, (25) is the average value and amplitude calculation unit, (26) is the average value, (27) is the average value and amplitude, (28) is the average value separation normalization unit, and (30) is the average value and amplitude. , (29) is a normalized vector, (31) is a codebook that stores the output vector, (32) is an output vector, and (33) is the normalized vector (2
9) and the distortion calculation section for calculating the distortion between the output vector (32), (2
2) is a threshold value, (34) is a block identification unit, (36) is the distortion obtained by the distortion calculation unit (33), and (37) is the minimum for detecting the minimum one from the distortion (36). The distortion detector, (35) is the block identification information, and (38) is the output vector index that gives the minimum distortion.

Next, the operation will be described. The input image signal sequence (1) is
The inter-frame prediction signal (3) is subtracted by the subtractor (2) and converted into the inter-frame difference signal (4). Since the inter-frame difference signal has smaller power than the original signal, it is a signal that can be encoded with a small encoding error.
The inter-frame difference signal (4) is encoded by the first stage vector quantization encoding unit (5). At this time, the threshold (22)
Is used as a parameter. The code data (6) coded in the first-stage vector quantization coding unit (5) is decoded in the first-stage vector quantization decoding unit (7) to obtain a first-stage decoded signal (8). A subtracter (9) subtracts the first-stage decoded signal (8) from the inter-frame difference signal (4) to obtain a first-stage error signal (10). Since the power of the first-stage error signal (10) is smaller than that of the inter-frame difference signal (4), it is possible to perform coding with a small coding error. The first stage error signal (10) is encoded in the second stage vector quantization encoding unit (11). At this time, the threshold value (22) is used as a parameter. The second stage vector quantization coding unit (1
The coded data (12) coded in 1) is decoded in the second stage vector quantization decoding unit (13) to obtain the second stage decoded signal (14). In the adder (15), the first-stage decoded signal
(8) is added to the second stage decoded signal (14) to obtain a decoded inter-frame difference signal (16). In the adder (17), the inter-frame prediction signal (3) and the decoded inter-frame difference signal (16)
And are added to obtain the decoded image signal sequence (18). The decoded image signal sequence (18) is stored in the temporary frame memory (19) and given a frame delay to form an inter-frame prediction signal. on the other hand,
The first-stage encoded data (6) and the second-stage encoded data (1
2) is variable-length coded in the variable-length coding unit (20),
It is temporarily stored in the buffer (21), subjected to speed smoothing processing, and then sent out as a transmission signal (24) via the line I / F 23. Further, the buffer (21) outputs it to a threshold value (22) proportional to the storage amount of the variable-length coded data and outputs it to the first stage vector quantization coding unit (5) and the second stage vector quantization. It is given to the encoding unit (11) to control the amount of information generated.

Next, along with FIG. 2 and FIG. 3, the first stage vector quantization coding unit (5) and the second stage vector quantization coding unit
The operation (11) will be described. As described for the conventional inter-frame vector quantizer, in vector quantization, a plurality of pixels are blocked and the output vector with the highest degree of similarity (low distortion) is searched. FIG. 2 shows an example of a vector quantized by the first stage vector quantization coding unit (5) and a vector quantized by the second stage vector quantization coding unit (11). Now the vector dimensions are all 4x
4 = 16. According to the present invention, a vector to be quantized in the first stage is formed by using an average value of the vector to be quantized in the second stage (formed from the first stage error signal) and the inter-frame difference signal corresponding to the same position on the screen as one element. To do. Therefore, as shown in Fig. 2, when the vector to be quantized in the first stage and the vector to be quantized in the second stage are superimposed on the screen, one in the first stage corresponds to 16 in the second stage. To do. In terms of the equation, the sequence obtained by blocking 4 × 4 = 16 inter-frame difference signals (4) is S _i = [S _i 1, S _i2 , ...,
S _i16 ], and a set of 4 × 4 = 16 blocks of this is [S] = [S ₁ , S ₂ , ..., S ₁₆ ] (S _i ∈
When expressed as [S], i = 1 to 16), FIG. 2 is expressed as follows.

Block average Vector quantized in the first stage m = [m ₁ , m ₂ , ..., M ₁₆ ]
m and [S] = [S ₁ , S ₂ , ..., S ₁₆ ] occupy the same position on the screen.

S ′ = the first-stage decoded signal (8) obtained by the first-stage vector quantization coding unit (5) and the first-stage vector quantization decoding unit (7).
[M ′ ₁ , m ′ ₂ , ..., M ′ ₁₆ ]
The first stage error signal (10) to be vector-quantized in the second stage is the following vector.

S _i −m ′ _i · u = [S _i1 −m ′ _i , S _i2 −m ′ _i , ..., S _i16 −m ′ _i ] (u is a unit vector) In the second stage, The vector quantization is performed after the average separation normalization is performed as described for the inter-frame vector quantizer of. Therefore, the influence of the first-stage vector quantization on the second-stage vector quantization is not related to the vector because it affects only the average value that is a scalar quantity. If there is an influence between the two, the efficiency of vector quantization in the second stage decreases due to the vector quantization error in the next stage. As in the present invention, if the relation between the first stage and the second stage is made to be a scalar quantity, the first stage vector quantization will not adversely affect the second stage vector quantization.

FIG. 3 is a block diagram showing a configuration example of the first stage vector quantization coding unit (5). The input signal to be vector-quantized is the inter-frame difference signal (4). The average value and amplitude of the signal (4) are calculated in the average value and amplitude calculating section (25) in units of 4 × 4 = 16 blocks. This block is the size of the block that is vector-quantized in the second stage. The obtained average value (26) is blocked to form a vector for the first-stage vector quantization, and the average value separation normalization unit (28) normalizes the average value separation. If you write in the formula, the calculation in the calculation unit (25) of the average value and the amplitude, And the processing in the average value separation normalization unit (28) is Is to ask. The normalized vector X = [x ₁ , x ₂ , ..., X ₁₆ ] obtained as described above and the codebook (3
The distortion with the output vector (32) read from 1) is calculated by the distortion calculator (33).
Ask in. Prior to this, the average value m _i and the amplitude g _i (27) are calculated in the block discriminator (34) by the threshold value (2
Compare with 2). Block identification is expressed as follows, where the threshold (22) is Th.

│m _i │ <Th and g _i <Th ...... invalid block │m _i │> Th or g _i > Th ...... valid block For the invalid block, neither the first stage nor the second stage needs to be encoded. However, since the block identification is based on the block in the second stage vector quantization coding unit (11), there is only one element for the first stage block. Therefore,
The distortion calculation unit (33) reduces the original distortion corresponding to the invalid block to 0.
And calculate. That is, the distortion (36) is usually However, if the block S _i is an invalid block, the corresponding element is S _i → m _i → x _i. And calculate. In the minimum distortion detector (37), the output vector and the input vector (29) stored in the code block 31 are stored.
The smallest of the distortions (36) with and (36) is detected, and the index number (38) of the output vector that gives the minimum distortion is output. The encoded data (6) output from the first-stage vector quantization encoding unit (5) is block identification information (35) and average value M.
And amplitude G (30) and index (38). However, if all the blocks to be quantized at the first stage are invalid blocks, information other than the block identification information (35) has no meaning and need not be transmitted, so the threshold (22) is set. Therefore, the amount of information generated can be controlled.

The operation of the second-stage vector quantization coding unit (11) is almost the same as the operation of the vector quantization coding unit (39) described for the conventional interframe vector quantizer. In the present invention, the threshold value (22) given to the second stage vector quantization coding unit (11) is controlled so that the first stage vector quantization coding unit (5)
Since only the coding can be used for coding, it is possible to significantly control the amount of information generated.

In the above embodiment, the threshold control is explained as the feed back control according to the buffer storage amount. However, the feed forward control of the feed forward control is performed by placing a frame memory in the preceding stage of the encoder to estimate the motion amount. It may be useful to add elements.

In the above embodiment, the fixed output vector set is used for the vector quantization codebook in both the first stage and the second stage, but the codebook for the first stage uses the frame memory contents in the loop. It is also effective to use a dynamic vector quantization or motion compensation method in which an output vector is generated. The merit that the average value of the blocks for the second-stage encoding is treated like pixels in the first stage is not limited to vector quantization. For example, even if orthogonal transform is used for the second-stage encoding, the error in the first stage is This is effective because it does not generate harmonics.

〔The invention's effect〕

As described above, according to the present invention, the vector quantizer has a multi-stage configuration, and the first-stage vector quantization is performed by using the average value of the block, which is the encoding unit in the latter stage, as the constituent element. There is an effect that the control range of the amount is wide, the image quality is not extremely deteriorated by the control, and the vector quantization error in the first stage does not adversely affect the encoding in the second stage.

[Brief description of drawings]

FIG. 1 is a block diagram showing the structure of an interframe vector quantizer encoder according to an embodiment of the present invention, and FIG. 2 is a first stage vector in an interframe vector quantizer encoder according to one embodiment of the present invention. FIG. 3 is an explanatory view showing the relationship between the quantization and the second stage vector quantization, and FIG. 3 shows an example of the configuration of the first stage vector quantization encoding unit in the interframe vector quantizer encoder according to the embodiment of the present invention. Block diagram, FIG. 4 is a block diagram showing a configuration example of a conventional interframe vector quantizer encoder, and FIG. 5 is a configuration of a vector quantization encoder in a conventional interframe vector quantizer encoder. It is a block diagram which shows an example. In the figure, (1) is the input image signal sequence, (2) is the subtractor, (3) is the inter-frame prediction signal, (4) is the inter-frame difference signal, (5) is the first stage vector quantization coding unit, ( 6) is the first-stage encoded data,
(7) is the first stage vector quantization decoding unit, (8) is the first stage decoded signal, (9) is the subtractor, (10) is the first stage error signal, (11) is the second stage vector quantization encoding unit, ( 12) is the second stage encoded data,
(13) is a second stage vector quantization decoding unit, (14) is a second stage decoded signal, (15) is an adder, (16) is a decoded interframe difference signal, (17) is an adder, (18) ) Is a decoded image signal sequence, (19) is a frame memory, (20) is a variable length coding unit, (21) is a buffer, (22) is a threshold, (23) is a line I / F, (24) Is the transmitted signal, (25) is the average value and amplitude calculation unit, (26) is the average value, and (2
7) is the average value and amplitude, (28) is the average value separation normalization part, (29) is the normalized vector, (30) is the average value and amplitude, (31) is the codebook, (32) is the output vector, (33) is the distortion calculation unit, (34) is the block identification unit, (35) is the block identification information, (36) is the distortion, and (3)
7) is the minimum distortion detection unit, (38) is the vector quantization index, (39) is the vector quantization coding unit, (40) is the coded data, and (41) is the vector quantization decoding unit. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (4)

[Claims] 1. A frame memory capable of storing at least one frame of image signal, and an inter-frame difference signal obtained by subtracting an inter-frame prediction signal read from the frame memory from an input moving image signal. A subtractor to be formed and an average value for calculating a first average value and an amplitude by blocking the inter-frame difference signal obtained by the subtractor into horizontal m pixels and vertical n lines (m and n are positive integers) An amplitude calculation circuit; a first-stage block identification circuit that compares the first average value and the amplitude with a first-stage threshold value;
The average value of m is divided into m horizontal blocks and n vertical blocks to include m × n blocks of the inter-frame difference signal,
A mean value separation and normalization circuit of the first stage for calculating the mean value and the amplitude of the mean value and subtracting the second mean value from the blocked first mean value to normalize by the second amplitude; and the mean value. A first-stage code table that stores a set of output vectors that match the distribution of input vectors with the first average value that has been separated and normalized as the first-stage input vector, and the distortion distance between the first-stage input vector and the output vector A first-stage distortion calculating circuit for calculating, and a first-stage minimum output detecting circuit for determining an index of the first-stage optimum output vector and the first-stage optimum output vector that gives the smallest distortion among the distortions calculated by the first-stage distortion calculating circuit A first-stage decoding circuit that obtains a first-stage decoded vector by multiplying the first-stage optimum output vector by the first amplitude and adding the second average value; and the inter-frame difference And a horizontal m-pixel vertical n-line so as to match the position on the screen where the inter-frame difference signal is blocked in the average value amplitude calculation circuit. The error signal is divided into blocks and the third average value and the amplitude of the other blocks except the block in which the first average value and the amplitude are smaller than the threshold value of the first stage in the block identification circuit of the first stage. And a third average value separation and normalization circuit that subtracts the third average value from the blocked error signal and normalizes it with a third amplitude, and the third average value and amplitude From the block identification circuit of the next stage to be compared with the threshold value of the stage, and the average signal separation and normalized error signal, the third average value and the amplitude of the next stage are determined by the block identification circuit of the next stage. A code table for the next stage that stores a set of output vectors that match the distribution of the input vector for the next stage, excluding blocks smaller than a certain value, and the input vector for the next stage and the output vector for the next stage. And a next-stage optimum output vector that gives the smallest distortion among the distortions calculated by the next-stage distortion calculation circuit and an index of the next-stage optimum output vector And a next-stage decoding for obtaining the next-stage decoded vector by multiplying the next-stage optimum output vector by the third amplitude and adding the third average value. A circuit, an adder for adding the decoded vector of the next stage to the decoded vector of the first stage to form a decoded interframe difference signal, and an adder for adding the decoded interframe difference signal to the interframe prediction signal An adder for obtaining a decoded image signal, a variable length code with the second and third average values and amplitudes, the optimum output vector indexes of the first and second stages, and the comparison result in the first and second block identification circuits as encoded data. And a buffer for temporarily storing the variable-length coded data to smooth the amount of information, wherein the first average value and the amplitude are If it is smaller than the threshold value of the first stage, the decoding vector of the first stage and the next stage of the corresponding block is set as a zero vector, and the third average value and the amplitude are set to the threshold value of the next stage in the block identification circuit of the next stage. If it is smaller, the decoding vector in the next stage of the corresponding block is set to be a zero vector, and the accumulated amount in the buffer is set as at least one control condition. Interframe vector quantizer hierarchical multistage, wherein the hierarchically control the stage and the next stage of the threshold. 2. When the distortion distance between the input vector of the initial stage and the output vector read from the codebook of the initial stage is obtained in the distortion calculation circuit of the initial stage, the first average is compared by comparison of the block identification circuit of the initial stage. 2. The interframe vector quantizer according to claim 1, wherein the element of the input vector of the first stage corresponding to the block whose value and amplitude are lower than the threshold value of the first stage is excluded from the distortion calculation. 3. A next-stage average value and amplitude quantization having a plurality of quantization characteristics, which is previously quantized by the variable-length encoding unit before variable-length encoding the third average value and amplitude. And the second value obtained by the average separation normalization circuit of the first stage.
The inter-frame vector quantizer according to claim 1, wherein the plurality of quantization characteristics are adaptively switched according to the amplitude of the. 4. A set of a plurality of output vectors having different characteristics is stored in the code table of the next stage, and the plurality of quantum values are stored by the second amplitude obtained by the average value separation and normalization circuit of the first stage. The inter-frame vector quantizer according to claim 1, wherein the quantization characteristic is adaptively switched.