JPH0549147B2 - - Google Patents

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention provides highly efficient encoding of an image signal sequence,
The present invention relates to an image signal sequential reproduction and encoding device for transmission and recording.

[Conventional technology]

Figure 5 shows, for example, the Institute of Electronics and Communication Engineers technical report IT85.
FIG. 61 (1985) is a block diagram showing an example of the configuration of a transmitting section of a conventional image signal sequential reproduction encoding device using a vector quantizer described in ``High-efficiency image encoding using vector quantization''. .

In Fig. 5, 1 is an input signal vector obtained by blocking an input image signal sequence such as still image data every m x n samples (m and n are natural numbers), and 2 is a previous stage that is delayed by one frame from input signal vector 1. A subtractor for obtaining a residual signal vector with the decoded vector 19, 3 is a residual signal vector, 6 is an average value within a block is obtained for each vector, the average value is subjected to high-efficiency encoding decoding, and average value encoded data 8 is obtained. and an average decoded value 9; 7 converts the residual signal vector 3 into a normalized output vector and an amplitude gain through a built-in product vector quantizer;
A normalized vector quantization encoder/decoder that outputs vector quantization encoded data 10 and amplitude reproduction output vector 11, 8 is average value encoded data, 9 is average value decoded value, 10 is vector quantization encoder 11 is an amplitude reproduction vector, 12 is a first adder which adds the average value decoded value to the amplitude reproduction vector to obtain a residual signal decoding vector 15, 15 is a residual signal decoding vector, 16 is the residual signal decoding A second adder adds the vector and the previous stage decoded vector 19 delayed by one frame to obtain a decoded vector 17 at each stage, 17 is a decoded vector at each stage, and 18 is for delaying the decoded vector by one frame time. 19 is a pre-decoded vector output from the frame memory, 20 is the average value encoded data 8 and vector quantized encoded data 1.
24 is encoded output data.

First, the principle of vector quantization will be explained very briefly. Now, the input signal sequence is K (=m×n)
Collectively, the input vector x = {x ₁ , x ₂ , ...,
x _K }. At this time, a set of N representative points (i.e., output vectors) y _i ={ y _i1 , y _{i2 .} . . , y _iK } of ^the K-dimensional Euclidean signal space R ^K ( y ₁ , y ₂ , ..., y _N ].The vector quantizer selects from among the set of output vectors:
The output vector y _i that is the shortest distance from the input vector (resulting in the minimum distortion) is determined as follows,
Explore this.

_if d( x , yi ) _< d( x , yl ) for all lx _{→ yi} where d( x , yi ) is the distance (distortion) between the input and output vectors. At this time, the input vector x is
It is transmitted or recorded according to the output vector index i, and is replaced by y _i during playback.

Mean-separated normalized vector quantization limits the distribution of output vectors to a unit hypersphere in a multidimensional signal space by mean-separated and normalized input vectors of a vector quantizer. now,
When the input signal vector is S = {S ₁ , S ₂ , ..., S _K }, the average value μ, the amplitude σ, and the average value separation normalized input vector x can be expressed as follows.

μ=1/K _K 〓 ^l=1 S _l σ=[1/K _K 〓 ^l=1 (S _l −μ) ² ] ^1/2 x _l = (S _l −μ)/σ x = {x _{1 ,} ^x _{2 , . _ _}

By scalar quantizing the average value μ and amplitude σ obtained in this way, and vector quantizing the average value separation input vector x , the versatility of the vector quantizer with a limited number of output vectors is increased, and the encoding performance will improve.

Next, the operation will be explained. In Figure 5,
First, during the first stage encoding, the frame memory 18 is kept in a cleared state. The input signal vector 1 is not processed by the subtracter 2 and is output as is as a residual signal vector 3. In the average value encoding decoder, after determining the average value of the residual signal vector 3, the average value is highly efficiently encoded using a technique such as DPCM encoding, and the average value encoded data 8 and local decoding are processed. The average decoded value 9 is output.
On the other hand, in normal vector quantization, the residual signal vector 3 is converted into a normalized output vector and an amplitude gain through an inner product vector quantizer, and vector quantized encoded data 10 and amplitude reproduction output vector 11 are converted into a normalized output vector and an amplitude gain.
Outputs .

Here, the operating principle of the inner product vector quantizer is explained in the sixth section.
This will be explained using figures. In the inner product vector quantizer, the normalized output vector y _i is average-separated and normalized and placed on the unit hypersphere of the multidimensional signal space.
The set Y=[ y ₁ , y ₂ , ..., y _N ] is used. That is, the normalized output vector y _i simultaneously satisfies the following conditions: | y _i |=[ _K 〓 ^j=l y _ij ² ] ^1/2 =1, _K 〓 ^j=1 y _ij =0.

The input vector x to the inner product vector quantizer is
Through the calculation process of the following equation, the input vector x is converted into a normalized output vector y _i that gives the maximum inner product value, and the maximum inner product value is given as the amplitude gain g of the input vector x .

if I ( x , y _i ) > I ( x , y _l ) for all l ( I ( x
, y _i )= _K 〓 ^j=1 (x _j y _ij )) x → y _i g=I ( x , y _i ) = | x | | y _i | cosθ _i = | x | cosθ _i is found as a scalar quantity The amplitude gain g is independently encoded and decoded, and the encoded data is output as vector quantized encoded data 10 together with the index i of the normalized output vector y _i , and at the same time, the amplitude gain g is added to the normalized output vector y _i . By multiplying by the locally decoded value g of the gain g, an amplitude reproduction output vector y _i ^* 11 is obtained.

y _i ^* =g・y _i After obtaining the residual signal decoded vector 15 by adding the average decoded value 9 and the amplitude reproduction output vector 11 in the first adder 12, the residual signal The decoded vector 15 is converted to 1 by the second adder 16.
It is added to the frame-delayed preceding and following vectors 19. The decoded vector 17 of each stage, which is the output of the second adder, is written into the frame memory 18 and delayed by one frame. On the other hand, the average value encoded data 8 and the vector quantization encoded data 10 are
The second code word is converted into an appropriate code word by the code assignment circuit 20 and sent out as encoded output data 24.
After the stage, input signal vector 1 and frame memory 1
The above encoding process is sequentially and repeatedly performed on the residual signal vector 3 with the previous-stage decoded vector 19 stored in 8.

[Problems to be Solved by the Invention] Since the conventional image signal sequential reproduction encoding device is configured as described above, the block size in each stage of encoding is always constant, and when the block size is increased, Problems include that the calculation scale of the inner product vector quantizer increases in proportion to the number of dimensions of the vector, and when the block size is reduced, it is difficult to significantly reduce the amount of encoded information generated in the first stage. It was hot.

This invention was made to solve the above-mentioned problems, and it is possible to make the block size variable without increasing the scale of calculations mentioned above, and to control the amount of encoded information in each stage over a wide range. The object of the present invention is to obtain an image signal sequential reproduction encoding device that can be executed.

[Means for solving problems]

The image signal sequential reproduction encoding device according to the present invention sets the number of dimensions of the input vector to L (=am×bn),
A vector/subvector converter that forms a subvector reduced from (a, b, m, n are natural numbers) to K (m x n), and a The number of dimensions L (=am×
subvector to interpolate and reproduce the vector of bn) /
A vector converter, and an encoding control unit that generates a control signal for adaptively switching the number of dimensions of the input vector and subvectors according to the number of encoding stages, and encoding/decoding processing from the second stage onwards. It is something.

[Effect]

In the image signal sequential reproduction encoding device according to the present invention, the number of dimensions of the vector in the vector/subvector converter and the subvector/vector converter and the encoding/decoding processing from the second stage onwards are controlled by the encoding control unit at each stage. adaptively controlled.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a transmitting section of an image signal sequential reproduction encoding apparatus according to an embodiment of the present invention. In the figure, 4 forms a subvector in which the number of dimensions of the input vector is reduced from L (=am×bn) (a, b, m, and n are natural numbers) to K (=m×n).

A vector/subvector converter 14 is a subvector/vector converter that interpolates and reproduces a vector of dimension L (=am×bn) using the sample values in the subvector of dimension K (=m×n); 5 is a residual signal subvector, 13 is a residual signal decoding subvector, and 22 is a parameter a, b, m, n and a threshold value that defines the number of dimensions L and K of the vector according to the number of encoding stages 21. control signal 2
The encoding control unit 25 that generates the residual signal vector 3 is identification information indicating whether to perform encoding/decoding on the residual signal vector 3. Also, 1-3, 6-12,
14 to 20 and 24 are completely the same as the conventional device.

Next, the operation will be explained. In Figure 1,
First, during the first stage encoding, the frame memory 18 is kept in a cleared state. Number of dimensions L=(am×
The input signal vector 1 of bn) is not processed by the subtracter and is output as is as the residual signal vector 3. The residual signal vector 3 is input to a vector/subvector converter 4, and is converted into a K (=m×n)-dimensional residual signal subvector 5 through the conversion process of the number of dimensions shown below, and is output. Ru. In the vector/subvector converter 4, the residual signal vector 3 and the residual signal subvector 5 are respectively input vectors; s = [s ₁ , s ₂ , ..., s _r , ..., s _L ]
:r(-{1, 2..., L} subvector; u = [u ₁ , u ₂ ,..., u _j ,..., u _K ]
: j (-{1, 2..., K}. The sample value u _j in the subvector is given by the averaged value of the samples in the input vector for each a×b sample, as shown in the following equation. It will be done.

u _j =1/a・b _r2 〓 ^r=r ₁ s _r r ₁ = (j-1)×a・b+1 r ₂ =j×a・b In Figure 2, a=b=m=n=4 The relationship between input vectors and subvectors in the case of is illustrated in correspondence with a two-dimensional image signal sequence.

The residual signal subvector 5 is input to a mean value encoder 6 and a normalized vector quantization encoder 7, and is subjected to the same processing as conventional ones. As a result, the average value encoded data 8 and the average value decoded value 9 are output from the average value encoder 6, and the vector quantization encoded data 1 is output from the normalized vector quantization encoder 7.
0 and the amplitude reproduction output vector 11 are respectively output, and the average value decoded value 9 and the amplitude reproduction output vector 11 are added in a first adder 12,
A residual signal decoding subvector 13 is obtained. The K-dimensional residual signal decoded sub-vector 13 undergoes the following processing through the sub-vector/vector converter 14 and is converted into an L-dimensional residual signal decoded vector 15. In the subvector/vector converter 14, the residual signal decoding subvector 13 and the residual signal decoding vector 15 are respectively converted into subvectors; u^ = [u ₁ ^, u ₂ , ..., u^ _j , ..., u^
_L ]: j(-{1, 2..., K} output vector; s^ = [s^ ₁ , s^ ₂ ,..., s^ _r ,..., s
^ _L ]: r (-{1, 2..., L}. The sample s^ _r in the output vector is the same sample in the subvector as a×b as shown in the following equation
It is given by repeating and arranging the samples one by one.

S^ _r = u _j :r(-{1, 2,...,L} j=INT{(r-1)/a・b} +1:INT(・) is converted to an integer by rounding down. Figure 3 shows a 2 is a two-dimensional diagram illustrating a sample array of subvectors corresponding to samples in an output vector when =b=m=n=4.

The residual signal decoded vector 15 is added to the previous stage decoded vector 19 delayed by one frame in a second adder in the same way as in the conventional one, and is written into the frame memory 18 as a decoded vector 17,
Used for the next stage of encoding. The average value encoded data 8 and the vector quantization encoded data 10
are respectively supplied to the code allocation circuit 20.

From the second stage onwards, the above encoding process is sequentially and repeatedly performed on the residual signal vector 3 between the input signal vector 1 and the previous stage decoded vector 19 stored in the frame memory 18 under the following control. The encoding control unit 22 controls the residual signal vector 3 in the encoding process according to the number of encoding stages 21;
Generate parameters a, b, m, and n that define the number of dimensions of each of the residual signal subvector 5, residual signal decoding subvector 13, and residual signal decoding vector 15, and a control signal 23 indicating a threshold value described later. The control signal 23 is then supplied to the vector/subvector converter 4, the subvector/vector converter 14, and the code assignment circuit 20. Figure 4 shows the number of dimensions L (=am
An example of a method for setting ×bn) will be shown in association with a two-dimensional image signal sequence. In addition, in the encoding process from the second stage onward, the vector/subvector converter 4
The root mean square value per sample in the residual signal vector 3 is compared in magnitude with the threshold value supplied from the encoding control section 22. As a result of the comparison, if the root mean square value is larger than the threshold,
The above encoding/decoding process is executed, and if not, the above encoding/decoding process is not performed and all sample values in the residual signal decoded vector 15 output from the subvector/vector converter are set to 0. do. The identification information 25 indicating the result of the size comparison is:
The signal is supplied from the vector/subvector converter 4 to the code assignment circuit 20 and the subvector/vector converter 14.

In the code allocation circuit 20, the control signal 2
3, the identification information 25, the average value encoded data 8, and the vector quantization encoded data 10;
are converted into appropriate code words and sent out as encoded output data 24.

In addition, in the above embodiment, in the operation of the vector/subvector converter, a method was shown in which a plurality of samples of the input vector are collectively averaged and used as a subvector sample. Samples extracted according to an appropriate subsample pattern may be used as subvector samples, and a smoothing filter for band limitation may also be inserted in the input stage.

〔Effect of the invention〕

As described above, according to the present invention, vector quantization can be performed hierarchically by reducing the number of dimensions of vectors to be encoded. Furthermore, since threshold processing is performed in the encoding process from the second stage onwards, the block size can be made variable without increasing the calculation scale of the inner product vector quantizer.
The amount of encoded information at each stage can be controlled over a wide range.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a transmitting section of an image signal sequential reproduction encoding device according to an embodiment of the present invention, and FIG. 2 is an explanation of an example of the operation of a vector/subvector converter according to an embodiment of the present invention. Figure, 3rd
FIG. 4 is an explanatory diagram showing an example of the operation of the sub-vector/vector converter according to an embodiment of the present invention, FIG. is a block diagram showing the configuration of a transmitting section of a conventional image signal sequential reproduction encoding device;
FIG. 6 is an explanatory diagram of the operating principle of the inner product vector quantizer. In the figure, 1 is an input signal vector, 2 is a subtracter, 3 is a residual signal vector, 4 is a vector/subvector converter, 5 is a residual signal subvector, 6
is an average value encoding decoder, 7 is a normalized vector quantization encoding decoder, 8 is average value encoding data,
9 is an average decoded value, 10 is vector quantization encoded data, 11 is an amplitude reproduction output vector, 12 is a first adder, 13 is a residual signal decoded subvector,
14 is a subvector/vector converter, 15 is a residual signal decoding vector, 16 is a second adder, 17
is a decoded vector, 18 is a frame memory, 19 is a previous stage decoded vector, 20 is a code allocation circuit, 21 is the number of encoding stages, 22 is an encoding control unit, 23 is a control signal, 24 is encoded output data, 25 is identification information It is. In addition, in the figures, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Scope of Claims] 1. A signal sequence obtained by delaying a reproduced image signal sequence obtained by performing high-efficiency encoding/decoding of the same input image signal sequence by one frame time, and the input signal sequence. In an image signal sequential reproduction encoding device that sequentially reproduces high-definition images as time passes by repeatedly performing high-efficiency encoding and decoding of the residual signal sequence again in frame units, the residual signal sequence is Grid-like difference signal series
L (= am x bn)-dimensional residual signal vector obtained by blocking each am sample x bn line (a, b, m, n are natural numbers, am = a x m, and bn = b x n) A vector/subvector converter that forms a K (=m×n)-dimensional residual signal subvector whose new sample sequence is the arithmetic mean value obtained for each grid-like a sample×b line. , an average value encoding decoder that performs high-efficiency encoding and decoding of the arithmetic mean value of the samples in the residual signal subvector;
dimensional input vector, vector quantizes the input vector into a K-dimensional normalized output vector with an average value of 0 and a size of 1 through a K-dimensional inner product operation, and converts the input vector into an inner product value of the normalized output vector and the input vector. a K-dimensional normalized vector quantization encoder decoder for highly efficient coding and decoding of the amplitude gain of the residual signal sub-vector given by the K-dimensional normalized vector quantization coding decoder for multiplying the normalized output vector by the decoded value of the amplitude gain An L-dimensional residual signal is obtained by arranging the samples in the K-dimensional residual signal decoding subvector obtained by adding the decoded value of the average value to each element of the dimensional amplitude reproduction output vector in a lattice shape by a sample x b line. and a sub-vector/vector converter that interpolates and reproduces the decoded vector, and the vector/sub-vector converter repeats the encoding/decoding operation while gradually reducing the size of the lattice block of the residual signal sequence. An image signal sequential reproduction encoding device characterized in that: 2. In the first-stage encoding, the vector/subvector converter takes the average value of a block of the input image signal sequence into grid-like a samples×b lines as one sample, and further converts the sample into a grid-like m×b line. A K-dimensional input vector is formed by grouping n samples each, and at the time of next-stage encoding, a K-dimensional input vector is formed by grouping the residual signal sequence by m samples x n lines in a grid pattern. The image signal sequential reproduction encoding device according to item 1. 3. At the time of the next stage encoding, a comparison is made for each block between the root mean square value of the sample values within the block, which is obtained by grouping the residual signal sequence into m samples x n lines in a grid pattern, and a predetermined threshold; The vector/subvector converter outputs only blocks whose root mean square value is greater than or equal to the threshold; and a code that highly efficiently encodes and decodes a binary signal sequence based on the magnitude comparison results of the vector/subvector converter. 3. An image signal sequential reproduction encoding apparatus according to claim 1, further comprising an allocation circuit.