KR910009092B1 - Vector quantum cooling and decoding apparatus - Google Patents

Abstract

No content.

Description

Device for vector quantization encoding and decoding between frames

1 is a block diagram showing the configuration of a vector quantization encoding and decoding unit of an interframe vector quantization encoding and decoding apparatus according to Embodiment 1 of the present invention.

2 is an explanatory diagram used for explaining the operation of the vector quantization encoding and decoding unit of the inter-frame vector quantization encoding and decoding apparatus according to the first embodiment of the present invention.

3 is a block diagram of a vector quantization encoding and decoding apparatus between frames of a multi-stage vector quantization structure according to another example of Embodiment 1 of the present invention.

FIG. 4 is a block diagram showing the structure of an ultra-short vector quantization encoding and decoding unit in the inter-frame vector quantization encoding and decoding apparatus of FIG.

FIG. 5A is an explanatory diagram showing the size of an ultra-short block when performing multistage vector quantization in the embodiment of FIG. 4. FIG.

FIG. 5B is an explanatory diagram showing the size of a block when performing multistage vector quantization in the embodiment of FIG. 4. FIG.

6 is a block diagram showing the construction of a vector quantizer according to Embodiment 2 of the present invention.

FIG. 7 is a diagram showing an example of the configuration of a codebook used in the vector quantizer of FIG.

8 is a block diagram showing another example of Embodiment 2 of the present invention;

FIG. 9 is a block diagram showing the construction of an inter-frame vector quantizer according to Embodiment 3 of the present invention. FIG.

FIG. 10 is a block diagram showing an example of the configuration of the band dividing section in FIG.

FIG. 11 is a block diagram showing an example of the configuration of a dynamic vector quantization coding unit in FIG.

FIG. 12 is a block diagram showing an example of a configuration of a dynamic vector quantization decoder in FIG.

FIG. 13 is a block diagram showing an example of the configuration of the codebook shown in FIGS.

FIG. 14 is a block diagram showing an example of the configuration of an encoding control section in FIG.

FIG. 15 is a diagram showing a relationship between an input amount of information generated and a threshold value output in the block identification threshold table and the distortion threshold table shown in FIG.

FIG. 16 is a block diagram showing the construction of a vector quantizer according to Embodiment 4 of the present invention. FIG.

FIG. 17 is a flowchart for explaining the encoding algorithm of FIG. 16. FIG.

18 is an explanatory diagram of operations of reading and writing quantization vectors with respect to the variable codebook of FIG.

19 is a flowchart for explaining a decoding algorithm of FIG.

20 is a block diagram showing the construction of a vector quantizer according to a modification of Embodiment 4 of the present invention.

21 is a block diagram showing the construction of a vector quantizer according to Embodiment 5 of the present invention.

FIG. 22 is a diagram showing the contents of a table indicating the career of the index transmitted in Embodiment 5 of the present invention. FIG.

FIG. 23 is a block diagram showing the construction of a vector quantizer according to Embodiment 6 of the present invention. FIG.

24 is a block diagram according to the most suitable example of the picture coding transmission apparatus according to Embodiment 7 of the present invention.

FIG. 25 is a detailed explanatory diagram of a vector quantization encoder of the apparatus of FIG. 24; FIG.

FIG. 26 is an explanatory diagram showing pixel arrangement of an adaptive spatial filter. FIG.

27 is an explanatory diagram of an example of a smoothing characteristic control of an adaptive spatial filter.

Fig. 28 is a block diagram of a picture coding transmission apparatus according to another embodiment of Embodiment 7 of the present invention.

FIG. 29 is a block diagram showing an example of the configuration of a block identification unit of a coding control method according to Embodiment 8 of the present invention. FIG.

30 is a diagram showing a relationship between a threshold value and an amount of information generation.

FIG. 31 is a diagram showing a relationship between a threshold and a time (frame). FIG.

32 is a diagram showing an example of setting a region.

Fig. 33 is a diagram showing the flow of processing in the block identification unit.

34 is a diagram showing the configuration of a video communication terminal having a multimedia data transmission method according to Embodiment 9 of the present invention.

35 is a diagram showing a frame structure of a multimedia data transmission method according to Embodiment 9 of the present invention.

36 is a view for explaining bit stream transmission to a transmission path by a multimedia data transmission method according to Embodiment 9 of the present invention.

FIG. 37 is a diagram showing a frame structure of a multimedia data transmission method according to another example of Embodiment 9 of the present invention. FIG.

38 is a block diagram of a conventional vector quantization encoding and decoding apparatus between frames.

39 is a block diagram of a vector quantization encoding and decoding unit in the inter-frame vector quantization encoding and decoding apparatus of FIG.

40 is a block diagram showing the configuration of a conventional vector quantizer.

FIG. 41 is a diagram showing a configuration example of a codebook used in a conventional vector quantizer. FIG.

42 is a block diagram of a conventional picture coding transmission apparatus.

43 is a block diagram showing the structure of a block identification unit of a conventional inter-frame encoding apparatus.

44 is a diagram showing a frame configuration of a conventional multimedia data transmission method.

45 is a diagram showing the contents of the FAS of the conventional multimedia data transmission method.

46 is a diagram showing a multiframe configuration of a conventional multimedia data transmission method.

Fig. 47 is a view for explaining the state of bit string transmission to a transmission path by a conventional multimedia data transmission method.

* Explanation of symbols for main parts of the drawings

1: subtractor 2: vector quantization coding and decoding unit

3: adder 4: frame memory

5 variable length encoder 6 transmission buffer

7: coding controller 8: average value separator

9: dot product vector quantization coder 10,11,31,32,33 codebook

12: scalar quantizer 13: normalization circuit

19,20,36: Selector 21: dot product vector quantization decoder

22: scalar quantization decoder 23: average value adder

30: Ultra-short vector quantization coding and decoding unit

34: ultra-short vector quantizer 35: average value calculation unit

The present invention relates to a vector quantization device between frames, and more particularly, to a vector quantization encoding and decoding device between frames for data-compressing moving picture signals by vector quantizing difference signals between frames.

In addition, the present invention blocks the input image information to generate an input vector, selects a representative vector closest to the input vector from a codebook that stores a plurality of quantization representative vectors, which are patterns of the input vector, and indexes the representative vector. The present invention relates to an encoding transmission apparatus for vector quantization using data as transmission data.

The present invention also relates to a vector quantization apparatus for encoding digital signals with high efficiency and to an encoding control method in an inter-frame encoding apparatus for encoding the input image signal sequence with high efficiency.

The present invention also relates to an improvement in an image coding transmission apparatus used for television communication such as a television conference, a television telephone, and a transmission method of a multimedia terminal in which multiplexed multimedia data such as voice, video, and data are transmitted. Such a conventional vector quantization apparatus, as shown in FIG. 38, is, for example, the vector quantization encoding between frames described in 175, Murakami et al., ″ Coding simulation between vector quantization scheme frames, " There is a decoding device. In FIG. 38, reference numeral 1 denotes a subtractor for outputting the difference signal sequence 103 between frames by subtracting the input image signal sequence 101 and the predicted image signal sequence 102 between frames. ) Is a vector quantization encoding and decoding unit which inputs the difference signal sequence 103 between the frames and the encoding control parameter 109 to output the difference data sequence 105 between the encoded data 104 and the decoded frame.

In the same drawing, reference numeral 3 denotes an adder which adds the predictive image signal sequence 102 between the frames and the difference signal sequence 105 between the decoded frames, and outputs the decoded image signal sequence 106.

(4) is a frame memory which gives a delay of a frame to the decoded image signal sequence 106 to form a predictive image signal sequence between the frames.

(5) is a variable-length encoding unit for inputting the encoded data 104 and outputs variable-length encoded data 107, and (6) inputs the variable-length encoded data 107 to encode a coding control instruction signal 108 And a transmission buffer for speed smoothing outputting the transmission data 110.

(7) is an encoding control unit which inputs the encoding control instruction signal 108 and outputs an encoding control parameter 109 to the vector quantization encoding and decoding unit 2.

The vector quantization encoding and decoding unit 2 is configured as shown in FIG. In Fig. 39, reference numeral 24 denotes an average value separation normalization unit for inputting the difference signal sequence 103 between the frames and outputting the separated average value 201, amplitude shift 203, and normalization vector 301.

(10) is a read-only codebook which stores several normalized output vectors (211), (25) selects a normalized output vector (211) so that distortion of the normalized vector (301) is minimized, and an index of the normalized output vector (211). A vector quantization encoder that outputs 209.

Numeral 14 denotes a block identifier that outputs the block identification data 204 by inputting the coding control parameter 109, the average value 201, and the amplitude shift 203.

Reference numeral 15 denotes an average value encoder which inputs the encoding control parameter 109 and the average value 201 to output the average value encoded data 205.

Numeral 16 denotes an amplitude gain encoder that inputs the encoding control parameter 109 and the amplitude gain 203 to output the amplitude gain encoded data 206.

Reference numeral 17 denotes an average value decoder for inputting the block identification data 204 and the average value encoded data 205 to output the decoded average value 207.

Reference numeral 18 denotes an amplitude gain decoder for inputting the block identification data 204 and the amplitude gain coded data 206 to output the decoded amplitude gain 208.

The block identification data 204, the average value encoded data 205, the amplitude gain encoded data 206, and the index 209 become the encoded data 104 and are sent to the variable length encoder 5 of FIG.

Reference numeral 26 denotes a vector quantization decoder for inputting the normalized output vector 211 output from the index 209 and the codebook 10 to output the selected output vector 302, and the reference numeral 27 denotes the output vector 302. An amplitude reproduction average value adding unit for outputting the difference signal sequence 105 between the input and decoded frames.

Next, the operation will be described. The input image signal sequence 101 subtracts the predictive image signal sequence 102 between frames by the subtractor 1 and converts it into the difference signal sequence 103 between frames. Since the difference signal sequence 103 between the frames has a smaller power than that of the original signal, encoding with less coding error can be performed at the same coding amount.

The difference signal sequence 103 between these frames is encoded and decoded with high efficiency by the vector quantization encoding and decoding section 2 described below to obtain a difference signal sequence 105 between the encoded data 104 and the decoded frame.

The adder 3 adds the predictive image signal sequence 102 between the frames and the difference signal sequence 105 between the decoded frames to obtain a decoded image signal sequence 106.

The decoded image signal sequence 106 is stored in the frame memory 4, and is delayed by a predetermined frame time to form the predictive image signal sequence 102 between frames for the next frame encoding.

On the other hand, the encoded data 104 is converted into suitable variable-length coded data (coded words) by the variable-length encoding unit 5, stored in the transmission buffer 6, and then smoothed out as a transmission data 110 at a constant speed. It is sent out.

In addition, the transmission buffer 6 obtains the total sum of the code amounts for the frames as the encoding control instruction signal 108 (hereinafter referred to as the encoding information generation amount) and supplies it to the encoding control unit 7.

The encoding control unit 7 controls the encoding used by the vector quantization encoding and decoding unit 2 in accordance with the encoding information generation amount 108 and encoding load signals such as the encoding speed and the reproduction quality which are fixedly selected by the external instruction. The parameter 109 is appropriately controlled.

High-efficiency encoding and decoding operations in the vector quantization encoding and decoding unit 2 will be described with reference to FIG. The input signal to be subjected to vector quantization is a differential signal sequence 103 between frames. The difference signal sequence 103 between these frames is formed into blocks (vectors) by the average value separation normalization unit 24, and average value separation normalization processing is performed.

If the blocked input signal series is represented by ε = (ε ₁ , ε ₂ , ......, ε _K ) (k = m ₁ × m ₂ , m ₁ , m ₂ is a natural number), For example, it is described as follows.

Since all normalized vectors X = (x ₁ , x ₂ , ......, x _K ) 301 obtained through the above process are distributed on the hypersphere in units of the K-dimensional signal space, the input vector before the mean value separation normalization is obtained. The effect of improving vector quantization efficiency is obtained as compared with the case of directly vector quantizing S.

About this normalization vector X, several normalization output vectors y ₁ (211) defined as quantization representative points are stored in the codebook 10 in advance.

This normalized output vector y ₁ is

Normalized under the condition that The vector quantization coder 25 selects a normalized output vector y ₁ having a minimum distortion with the normalized vector x, and outputs an index i 209 for identifying the normalized output vector. That is, the following expression is executed.

On the other hand, the separated average value m 201 and the amplitude gain g 203 are each independently high efficiency by the average value encoder 15 and the amplitude gain encoder 16, respectively.

Coding characteristics such as the number of quantized bits, the quantization width, and the like of the scalar quantizer used in the average encoder 15 are appropriately controlled in accordance with the encoding control parameter 109.

The average value m 201 and amplitude gain g 203 are used by the block identification section 14 together with the coding control parameter 109 for block identification.

That is, the magnitude comparison with the threshold value Th corresponding to the encoding control parameter 109 is executed according to the following equation to determine the block identification data ν 204.

This effective block is output as the encoded data 104 together with the average value encoded data 205, the amplitude gain encoded data 206, the index 209, and the block identification data 204 corresponding to the block.

(207), 진폭이득

(208) 및 벡터 양자화 복호화기(26)에서 코드북(10)에서 리드된 상기 인덱스(209)에 대응하는 정규화 출력벡터 y₁(302)를 사용해서 진폭재생 평균값 가산부(27)에서 다음의 국부복호동작이 실행되어 복호 프레임간의 차분신호(105)로 되는 복호벡터

가 얻어진다.Moreover, the average value decoded through the average value decoder 17 and the amplitude gain decoder 18.

207 amplitude gain

208 and the next local in the amplitude reproduction average value adding unit 27 using the normalized output vector y ₁ 302 corresponding to the index 209 read from the codebook 10 in the vector quantization decoder 26. Decoding vector to perform decoding operation and become difference signal 105 between decoding frames

Is obtained.

For an invalid block, all the difference signals between the frames of the block are treated as zero. Therefore, the coded data 104 to be output may be the block identification data 204 alone, and the average value coded data 205, the amplitude gain coded data 206, and the index 209 need not be transmitted.

(207) 및 진폭이득

(208)을 함께 0으로서 출력하는 것에 의해 복호벡터

는In addition, the average value decoded in the average value decoder 17 and the amplitude gain decoder 18

207 and amplitude gain

Decoding vector by outputting 208 together as 0

Is

Is given.

Since the conventional vector quantization encoding and decoding apparatus for frames is configured as described above, it is difficult to efficiently perform adaptive encoding processing corresponding to the variation of the characteristics of the input image, and to reduce the block identification threshold value for improving the quality of the reproduced image. In this case, there was a problem such that the amount of information generated increases extremely.

In addition, the structure of the conventional vector quantization described in the Telecommunications Research Institute Technical Report TT85-61 is as shown in FIG. In Fig. 40, reference numeral 25 denotes a vector quantization encoder for vector quantizing the input vector 101, and outputs the index of the normalized representative vector as the coded data 113, code 10 denotes a codebook, code 26 denotes an encoding. A vector quantization decoder which reproduces the normalized representative vector corresponding to the index given in the data 113 as a decoded vector 114.

In the above-described vector quantization process of the vector quantizer, the tree search method described below is used for speeding up the operation during the normalized representative vector search. 41 is an example of a normalized representative vector arranged in a binary tree shape. It is designed in advance so that the upper vector becomes a representative vector of the lower vector.

In each stage, an operation of selecting one of two vectors whose distortion with the input vector X is smaller is sequentially performed from the top to the bottom to finally determine the normalized representative vector. In the case of a binary tree, ″ 0 ″ or ″ 1 ″ is allocated according to the branching direction at each node, and a binary sequence representing a path leading to the lowest normalized representative vector corresponds to the index i of the normalized representative vector. do.

Since the conventional vector quantizer is configured as described above, when the number of vectors is high, it is difficult to completely optimize the finite normalized representative vectors previously stored in the codebook for all the information source input vectors. There was a problem in that excessive normalization error should be reduced.

In addition, the conventional interframe vector quantizer must encode without distinguishing a visually sensitive input video signal sequence having a low intermediate frequency and a visually insensitive input image signal series having a high spatial frequency. There was a problem that was difficult.

A conventional picture coding transmission apparatus will be described with reference to the drawings.

Fig. 42 shows the block configuration of the picture coding transmission apparatus disclosed in " Image Quality Control in Moving Picture Coding, Atsuda Ito (1986 Picture Coding Symposium Preliminary Manuscript 6.3) ".

As shown in FIG. 42, this picture coding transmission apparatus is composed of a preceding processor A, a motion compensator B, a vector quantization encoder C and a vector quantization decoder D. As shown in FIG. Then, the front processor A reads the same image signal every frame and performs analog digital conversion (hereinafter referred to as A / D conversion) to generate an input image signal 101 and an image image. A block divider 92 is used to block pixels at positions approaching each other by a predetermined number and to generate an image vector signal 142 that becomes an input image signal 101 group for each block.

In addition, the motion compensating unit B has a current block position in the frame memory 4 storing the decoded reproduction signal 152 of the previous frame and the decoded reproduction signal 152 of the previous frame stored in the frame multimedia 4. And a motion compensation processing unit 93 for generating a plurality of reference blocks 143 on the basis of the?, And searching and outputting the reference block 144a and the motion position information 144b that are closest to the image vector signal 142. .

The vector quantization unit C and the vector quantization decoding unit D have the same structure as the conventional structure described above.

The spatial filter 94 is a filter for smoothing the decoded reproduction signal 151.

Next, the operation of the conventional picture coding transmission apparatus will be briefly described.

The motion image signal 100 of the first frame is input to the A / D converter 91, converted into an input image signal 101, and input to the block divider 92. Then, the image signal 142 is generated and output by arranging the signal 101 at the position approached on the image by the block divider 92 by a predetermined number.

On the other hand, the encoded vector signal 104 is converted from the vector quantization decoder 26 to the decoded vector signal 150, and is added to the reference block signal 144a by the adder 3 to the decoded reproduction vector signal 151. Is converted.

When the motion position information 144b is large, the decoded reproduction vector signal 151 is smoothed and stored in the frame memory 4. Here, the smoothing process of the spatial filter 94 is made in accordance with the motion position information 144b obtained by motion compensation to perform the smoothing process only for the same area, and not to perform the smoothing process for the still area.

Since the conventional picture coding transmission apparatus is configured as described above, the filter is smoothed by the filter even for a portion with high precision encoding that becomes a decoded picture close to the input motion picture signal. In order to perform ON / OFF, there is a problem that a smoothing process is not performed on a still region, and encoding noise of a poorly encoded region accumulates, resulting in deterioration of a decoded image.

43 is a block diagram of an inter-frame encoding apparatus using the conventional decoding control method described in " TV Conference Color Motion Picture Transmission Method " of, for example, Telecommunications Society Technical Report CS85-3, Murakami Ito, Kamisawa et al. It is a block diagram showing the structure of an identification part.

In FIG. 43, reference numeral 73 denotes a threshold controller which outputs a threshold Th 74 corresponding to the encoding control parameter 109, and 75 denotes a block for performing block determination by the threshold Th 74. In FIG. It is a judgment part.

The block identification unit 14 outputs Th 74 from the threshold control unit 73 so as to determine a block of the entire frame at the threshold Th corresponding to the encoding control parameter 109 in the encoding control unit. The comparison of the magnitudes of the data values is executed in the block determining unit 75 and the block identification data 204 is determined according to the following equation.

For this valid block, the data value corresponding to the block is output as coded data together with the block identification data v (204), and the data value corresponding to the block in the invalid block is regarded as 0, so that only the block identification data v (204) is encoded. It is output as data.

Since the encoding control method used in the conventional interframe encoding apparatus is executed as described above, if the block identification threshold is lowered in units of frames when the input image is stopped, the number of effective blocks in units of one frame increases and the amount of information generated is extremely high. There was a problem of increasing.

Fig. 44 shows, for example, the structure of the interface in a 64 kbps-based multimedia communication system and its application, Matsudo Juicy, 1987 Technical Report of the Institute of Electronic and Information Communication, and the frame structure of the conventional multimedia data transmission method described in ICS87-7. In the drawing, reference numeral 76 denotes an octet frame of eight sub-channels of 1 bit each having a repetition period of 8 KHz, and reference numeral 77 denotes a transmission composed of 80 eight octets 76. Frame.

FIG. 45 is a table for explaining assignment of the frame synchronization signal FAS in FIG.

FIG. 46 is a table showing the details of the frame synchronizing signal FAS in a multi-frame composed of 16 transport frames 77 in FIG.

FIG. 47 is a diagram showing a state in which the transmission frame shown in FIG. 45 is transmitted to the transmission path having a bit rate of 64 kbps. In the drawing, reference numeral 78 denotes a bit example transmitted to the 64 kbps transmission path.

Next, the operation will be described. In FIG. 44, the octets 76 have a length of 8 bits. The period TocT of the octet frame 76 in the case of sending this to the line having a transmission rate of 64 kbps is

And the transmission capacity Cs for each 1-bit sub channel is

It becomes In other words, when subchannels in the octets 76 are individually allocated to audio, video, data, and the like, each allocation rate is an integer multiple of 8 kbps. This has the advantage that matching with a speech coded decoding apparatus (voice codec) having a sampling frequency of 8 KHz is easy to take. The transmission frame 77 is an eighth sub-channel in the octet frame 76 for the purpose of identifying the sub-channels in this octet, 76, or the like to identify the allocation to each media to the subchannel. It is composed by collecting 80 octagonal frames 76 by frame bits occupying (service channel). The eighth sub-channel (service channel) in this transmission frame 77 is an 80 octagonal frame 76 and its use is simple.

The eighth sub-channel (service channel) is composed of an 8-bit frame synchronization signal FAS, an 8-bit bit rate allocation signal BAS, and a 64-bit application channel AC. In order to identify the frame synchronization and the multiframe synchronization formed by collecting 16 frames of the transmission frame 77, the content shown in FIG. 45 is arranged in the FAS as a frame synchronization pattern. The receiving side first detects the unique pattern to establish synchronization of the transmission frame 77. Next, the multi-frame synchronization is established based on the Mi bit shown in FIG. FIG. 46 is a table showing the contents of the Mi bits shown in FIG. 45, in which 16 bits are allocated as Mi bits in one multiframe, and unique patterns and additional information are arranged.

The BAS in the transmission frame 77 can dynamically change its allocation information for each of the eight transmission frames SMF1 and SMF2 77 each divided into two multiframes shown in FIG. 46, and each of the transmission frames 77 for SMF1 and SMF2. ), The same BAS information is sent consecutively once. The receiving side applies an error protection means for identifying the bit rate allocation in the next SMF based on the BAS information whose contents have been matched more than five times out of these eight times.

Application channel AC is allocated to command data for end-to-end negotiation at initial setting, but is allocated to user data during communication to effectively use the line. The bit capacity CAc of this AC unit is given by the following.

FIG. 47 shows the specific configuration of the transmission frame 77 transmitted on the 64 kbps line according to the above description.

This transmission frame configuration is based on bit rate allocation for each subchannel having a capacity of 8 kbps. For example, it is obvious that it cannot be applied to a 56 kbps line common in North America and a 32 kbps line commonly used in an enterprise communication network. Do.

Since the conventional multimedia data transmission method is configured as described above, it is difficult to apply to a line of l × 8 kbps (L is an integer of 1 or more). For example, in order to have a function suitable for a l × 8 kbps line with the same device, each transfer rate In response to the need to set individual frame formats, there is a problem that the amount of H / W increases and the transmission capacity cannot be used effectively.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above problems, and it is possible to suppress the waveform distortion between the input and output vectors in the vector change amount between frames to a predetermined value or less, and to generate the amount of encoded information by changing the threshold value for the waveform distortion. The present invention provides an apparatus for vector quantization encoding and decoding between frames, which can adaptively control the quality of a reproduced picture and a quality of a reproduced picture widely, and which can generate and update a codebook depending on a local property of an input picture.

Another object of the present invention is to provide an inter-frame vector quantization encoding and decoding apparatus capable of suppressing generation of an effective block in inter-frame vector quantization and reducing an amount of encoding information generation.

Another object of the present invention is to provide a vector quantizer capable of sufficiently reducing the quantization error even for an unusual input vector.

Another object of the present invention is to provide a quantizer between frames that can be efficiently encoded in consideration of visual characteristics and have a high subjective quality of an encoded reproduction picture.

It is still another object of the present invention to provide an image encoding and transmission device capable of effectively suppressing coding noise and at the same time suppressing the quality of a reproduced image locally.

It is still another object of the present invention to provide an encoding control method capable of suppressing a sudden increase in the number of effective blocks when the block identification threshold is lowered, thereby reducing the amount of information generated.

It is still another object of the present invention to provide a multimedia data transmission method capable of coping with various transmission speeds with the same H / W as a frame configuration applicable to a L × 8 kbps line as an integrated frame configuration.

In the inter-frame vector quantization encoding and decoding apparatus according to the present invention, in a vector quantization encoding and decoding unit of an inter-frame encoding loop, an inner vector quantization encoder having a plurality of codebooks or a vectorized differential signal sequence between frames is normalized and average-value separated and normalized. An average value separation normalization unit and a vector quantization encoder for processing and quantization are provided.

Further, in the present invention, the vector quantization encoding and decoding apparatus between frames vectorizes a differential signal sequence between frames to an output terminal of the frame memory and outputs an ultra-short vector quantization decoded signal sequence and an ultra-stage vector quantized coded data. We have made wealth.

The internal vector quantization coder of the vector quantization coding and decoding unit according to the present invention, when the waveform distortion calculated in the vector quantization coding process is larger than a threshold value, scalar quantizes and transmits an input vector having a mean value, and performs a normalization process. Are stored sequentially, and are used as an output vector for subsequent encoding, and the content of a codebook used for vector quantization with time is dynamically updated in accordance with the characteristics of the input image, or the difference signal series between frames is averaged. The vector is separated by a normalization unit, and the average value normalization is performed. The vector quantization encoder performs quantization of the normalized vector, and in this quantization process, the minimum distortion is extracted and the contents of the codebook are sequentially updated according to the value obtained by loading the amplitude gain. .

In the present invention, an ultra-short vector quantization encoding and decoding unit converts the input image signal sequence into an input vector, reads the output vectors from several codebooks using the inter-frame predictive image signal sequence in the frame memory, and vectorizes the input vector. If the minimum distortion obtained in the quantization process exceeds the quantization threshold, the first average value of the input vector is obtained and interpolated to update the contents of the codebook, and the interpolated first average value is output as the first vector quantization decoding sequence, and the first average value is added. Ultra-short vector quantized coded data is output in addition to the ultra-short index for output vector identification.

The vector quantizer according to the present invention is automatically generated by sequentially storing a sixth codebook which is a standard quantization representative vector and a unique input vector extracted according to the minimum distortion as a new quantization representative vector. The first and second codebooks are constructed in a tree shape while transmitting the unique input vector.

In the present invention, when the minimum distortion is larger than the set distortion threshold, the vector quantizer is transmitted with the input vector at that time, and is stored as the new quantization representative vector in the seventh codebook and used for subsequent vector quantization processing. Further, the search operation is performed at high speed by constructing the sixth and seventh codebooks in a tree shape.

The inter-frame vector quantizer according to the present invention includes a band divider for band-dividing an input image signal sequence, a frame memory for storing the input image signal sequence for each frequency band, and encoding and encoding according to characteristics of a predetermined frequency band input image signal sequence. A coding control unit for generating a control signal for switching the precision of the coding process is provided in accordance with the dynamic vector quantization coding unit for performing the decoding process, the dynamic vector quantization decoding unit, and the spatial frequency of the input image signal sequence.

In the present invention, the vector quantizer between frames stores an input image signal sequence for each frequency band generated by a band divider in a frame memory, and dynamically divides the input image signal sequence for each frequency band in the frame memory in time division. The dynamic vector quantization coder sends the vector quantization coder an encoding process of precision according to the spatial frequency of the input image signal sequence in accordance with a control signal supplied from the coding control unit.

Also, two vector quantizers according to the present invention use a fixed codebook to perform vector quantization on a quantized coder, and when the distortion is larger than an allowable value, the coder converts the image signal transmitted as unprocessed into a variable codebook. It has a quantization coder for vector quantization by using them, and these two vector quantization coders are connected in series.

In the present invention, when the minimum distortion is larger than the set distortion threshold, the input vector is not encoded at all by the preceding encoder, and is transmitted to the next vector quantization encoder. Here, the distortion estimation is performed, and here, when larger than the threshold value, the input vector is output to the decoding side as the original signal, but when smaller than the other threshold value, the index number of the representative vector giving the minimum distortion is output.

The vector quantizer according to the present invention transmits a unique input vector extracted according to the minimum distortion and rewrites and stores the most recently used quantization representative vector stored in the codebook as a new quantization representative vector. It is intended to be updated.

When the minimum distortion is larger than the set distortion threshold, the vector quantizer according to the present invention transmits the input vector and rewrites and stores it as the quantization vector most recently used in the codebook as a new quantization representative vector. Used for In addition, since index data whose code length is as short as the quantization representative vector having a high selection frequency is allocated according to the selection frequency, the quantization efficiency is improved.

An image encoding transmission device according to the present invention includes an encoding precision control unit for switching the encoding precision of the vector quantization encoding unit in a predetermined period according to the amount of transmission encoding information of an encoded image vector signal temporarily stored in a transmission buffer, and each pixel of the decoded reproduction signal. An adaptive spatial filter for smoothing the decoded reproduction signal and the smoothing process of the adaptive spatial filter are turned ON / OFF in accordance with the motion position information. And a smoothing characteristic control unit for controlling the smoothness of the adaptive spatial filter when the coding accuracy is low, and for reducing the smoothness when the coding precision is high.

Since the image encoding transmission apparatus of the present invention controls the spatial filter characteristic in the encoding loop between frames in accordance with the encoding precision in units of image blocks, encoding noise is effectively suppressed and the encoded image quality is improved. The encoding control method according to the present invention does not lower the block identification threshold value equally to the desired value in units of frames when the image is stopped, but gradually decreases the threshold value after a predetermined time in time and through the steps in space. In this case, the threshold value of the entire frame is lowered to the desired value while enlarging the area falling to the desired value.

The encoding control method according to the present invention can suppress a sudden increase in the number of effective blocks and an extreme increase in the amount of information generation by changing the block identification threshold in the frame.

The multimedia data transmission method according to the present invention enables the same transmission frame structure on the basis of a L bit length frame having a repetition period of 8 KHz in a L x 8 kbps line.

The above and other objects and novel features of the present invention will become apparent from the description and the accompanying drawings.

Example 1

Hereinafter, Embodiment 1 of the present invention will be described with reference to the drawings. In Fig. 1, reference numeral 8 denotes an average value divider which inputs a differential signal sequence 103 between frames and outputs an average value separated input vector 202 and an average value 201, and 9 denotes this average value separated input vector 202. ), A dot product vector quantization encoder that outputs a mean value-separated input vector 202 and an index 209 by inputting the normalized output vector 211 and the encoding control parameter 109.

10 denotes a first codebook identical to the conventional codebook 10, and 11 denotes a second codebook that can be written and read from time to time, respectively, and outputs a normalized output vector 211, and the second codebook 11 The normalized input vector 210 is input to the. Reference numeral 21 is a scalar quantizer that inputs the average value separation input vector 202 to output a scalar quantization value 212 and outputs a scalar quantized average value separation input vector 214 for each sample.

(13) inputs the average value separated input vector 214 outputted from the scalar quantizer 12 as a normalization circuit and outputs the normalized input vector 210 to the second codebook 11.

(19) inputs the index 209 output from the dot product vector quantization encoder 9 and the scalar quantization value 212 output from the scalar quantizer 12 as a first selector, and outputs the vector encoded data 213. do.

(20) inputs this vector coded data 213 as a second selector, and outputs an index 209 and a scalar quantized value 212.

Reference numeral 21 denotes a dot product vector quantization decoder for outputting a selected normalized output vector 216 by inputting an amplitude gain 208, an index 209, and a normalized output vector 211 output from the amplitude gain decoder 18. .

Reference numeral 22 denotes a scalar quantization decoder that inputs a scalar quantization value 212 output from the second selector 20 and outputs a scalar quantized average value separation input vector 214 for each sample.

Reference numeral 23 denotes an average value adder for inputting the normalized output vector 216 and the average value separated input vector 214 to output the difference signal sequence 105 between the decoded frames. The rest of the configuration is similar to that of FIG. 39, and FIG. 1 shows the structure of the vector quantization encoding and decoding unit of the inter-frame vector quantization encoding and decoding apparatus according to the present invention. And the decoding unit 2, and other portions are the same as those in FIG.

Next, the operation will be described. The difference signal sequence 103 between the frames supplied from the subtractor 1 shown in FIG. 38 is input to the average value separator 8 to be block (vector).

In this average value separator 8, the average value m 201 of the vectorized input signal sequence ε is separated and the average value separation input vector Z 202 is output. This process is represented by the following equation.

In the dot product vector quantization encoder 9, an inner product operation is performed on the average value separated input vector Z 202 and the normalized output vector y ₁ described in the first codebook 10 and the second codebook 11. The normalized output vector y ₁ with the maximum dot product is detected, and the maximum dot product at that time is approximately given as the amplitude gain g. In other words, the amplitude gain g 203 and the index i 209 are simultaneously obtained through the following equation.

The amplitude gain g is limited to a value of zero or more.

Here, the waveform distortion D by vector quantization is defined by using the magnitude | Z | of the mean-separated input vector Z and the amplitude gain g approximated by the maximum dot product.

FIG. 2 shows the relationship between the average value separated input vector Z, the normalized output vector y ₁ , the amplitude gain g and the waveform distortion D. The encoding process is divided into two cases according to the result of the magnitude comparison between the waveform distortion D and the allowable distortion threshold T _D specified by the encoding control parameter 109.

The amplitude gain g 203 and the index i 209 obtained in the above process are output as it is.

The amplitude gain g 203 is output as a negative constant value (for example, -1), and the average separated input vector Z 202 is supplied to the scalar quantizer 12.

(214)와 K개의 스칼라 양자화값(212)가 출력된다.The scalar quantizer 12 quantizes the average separated input vector Z 202 according to the quantization characteristics specified by the coding control parameter 109 for each sample, and the scalar quantized average separated input vector.

214 and K scalar quantization values 212 are output.

(214)는 정규화회로(13)에서 다음의 정규화 처리를 받아 정규화 입력벡터

(210)으로 변환된다.The scalar quantized mean value separated input vector

A normalization input vector is received by the normalization circuit 13 in step 214.

Is converted to 210.

(210)은 제2도의 코드북(11)의 소정의 어드레스상에 라이트되고, 이상의 내적 벡터 양자화 부호화 처리에서의 정규화 출력벡터 y₁로서 리드된다.This normalized input vector

Reference numeral 210 is written on the predetermined address of the codebook 11 in FIG. ₂ and read as the normalized output vector y ₁ in the dot product vector quantization coding process described above.

The predetermined address starts at address 0 and is controlled to be sequentially counted up according to the write operation and reset to address 0 at the time when the last address is exceeded.

The block identification unit 14, the average value encoder 15, the amplitude gain encoder 16, the average value decoder 17, and the amplitude gain decoder 18 perform the same operations as the conventional example shown in FIG.

The first selector 19 is input with the index 209 or the K scalar quantized values 212 and the amplitude gain 208 decoded by the amplitude gain decoder 18, and the decoded amplitude gain. When 208 is 0 or more, the index 209 is selected and when K is less than 0, the K scalar quantization values 212 are selected, respectively, and the vector encoded data 213 is output.

Therefore, as the encoded data 104, when the block identification data v 204 output from the block identification unit 14 indicates 1, that is, an effective block, the block identification data 204, the average value encoded data 205, and the amplitude gain are obtained. The encoded data 206 and the vector encoded data 213 are output.

When the block identification data v 204 indicates 0, that is, an invalid block, only this block identification data v 204 is output.

(208)과 상기 벡터 부호화 데이타(213)이 입력된다. 이 복호화된 진폭이득

(208)이 0이상일 때는 벡터 부호화 데이타(213)을 인덱스(209)로서 내적 벡터 양자화 복호화기(21)에 공급하고, 0미만일 때는 상기 K개의 스칼라 양자화값(212)로서 스칼라 양자화 복호화기(22)에 공급한다.The second selector 20 has a decoded amplitude gain

208 and the vector encoded data 213 are input. This decoded amplitude gain

When 208 is equal to or greater than zero, the vector encoded data 213 is supplied to the dot product vector quantization decoder 21 as an index 209, and when less than 0, the scalar quantization decoder 22 is used as the K scalar quantization values 212. Supplies).

(208)을 인덱스(209)에 대응해서 제1의 코드북(10)과 제2의 코드북(11)에서 리드되어서 선택된 정규화 출력벡터 y₁에 승산해서 진폭 재생된 출력벡터 gy₁(216)을 얻는다.Decoded amplitude gain in amplitude gain decoder 18 in dot product vector quantization decoder 21.

208 is multiplied by the first codebook 10 and the second codebook 11 corresponding to the index 209 to the selected normalized output vector y ₁ to obtain an amplitude reproduced output vector gy ₁ 216. .

(214)를 얻는다.In addition, the scalar quantization decoder 22 performs a scalar quantization decoding operation according to the quantization characteristics specified by the K scalar quantization values 212 and the coding control parameter 109, and performs a scalar quantized average value separated input vector.

Get (214).

(207)을 가산하는 것에 의해, 즉 다음식의 연산을 실행하는 것에 의해 복호 프레임간의 차분신호(105)로 되는 복호 벡터

가 구해진다.In the average value adder 23, the decoded average value supplied from the average value decoder 17 to the amplitude reproduced output vector gy ₁ 216 or the scalar quantized average value separated input vector Z 214.

A decoding vector that becomes a difference signal 105 between decoded frames by adding (207), i.e., performing the following equation.

Is obtained.

는When the block identification data ν 204 is 0, the decoding vector is performed in the same manner as in the prior art.

Is

Is given.

In this embodiment, waveform distortion between input and output vectors in inter-frame vector quantization can be suppressed to a predetermined value or less, and the amount of encoded information generation and reproduction quality can be widely applied by changing the threshold for waveform distortion. have.

It is also possible to generate and update codebooks depending on the local nature of the input image while coding.

In the above embodiment, the vector quantization encoding and decoding unit of the inter-frame vector quantization encoding and decoding apparatus has described a case where the inner product vector quantization is performed using a codebook sequentially updated according to the waveform distortion. The means for performing quantization and sequentially updating the codebook may be used in accordance with the value obtained by loading amplitude gain to the minimum distortion obtained in the quantization process. The same effect can be obtained by using a vector quantizer instead of the scalar quantizer.

In addition, at the output terminal of the frame memory 4, a plurality of output vectors which become block images cut out on a predetermined address of the frame memory 4, a plurality of output vectors which become uniform patterns of a predetermined level, and past input image signals A multi-stage vector quantization configuration may be provided with a quantization encoding and decoding unit that directly vector quantizes an input image signal sequence using a plurality of output vectors serving as average value patterns for several samples of the sequence.

3 is a block diagram of an inter-frame vector quantization encoding and decoding apparatus according to this multi-stage vector quantization configuration. In FIG. 3, reference numeral 30 denotes an ultra-short vector quantization encoding and decoding unit, which predicts an image signal sequence 102 between frames, which is inputted from the frame memory 4 to the input image signal sequence 101 together with the code encoding control parameter 109. Block) for each horizontal and vertical sample, and outputs the ultra-short vector quantized decoded signal sequence 121 to the subtractor 1 and the adder 3, and also the variable-length coder 5 as the ultra-short vector quantized coded data 120. It is supposed to output to. Other configurations are the same as those in FIG.

The detailed block structure of this ultra-short vector quantization encoding and decoding unit 30 is shown in FIG. 4, and in the drawing, reference numeral 34 denotes an input image signal sequence 101 and third, fourth and fifth codebooks. A plurality of output vectors read in (31) to (33) are input, and are an ultra-short vector quantizer for outputting an ultra-short index 122, an input image signal sequence 101, and an output vector 124.

The fourth codebook 32 stores a dynamic output vector obtained from several blocks cut out on a predetermined address of the frame memory 4 of FIG. 3 in the ultra-short vector quantizer 34, and then writes and reads at that time. Would be possible.

The third codebook 31 is read only for storing a plurality of fixed value output vectors having a uniform level.

The fifth codebook 33 can be written and read by storing the output vector 125 interpolated with a plurality of average values.

(35) is an average value calculating unit, when the minimum distortion Ds is larger than the ultra-short vector quantization threshold, the average value is obtained for each small block on the two-dimensional image sample array, and the ultra-short average value 123 is output. 125 is output to the fifth codebook 33 and the third selector 36.

(124)를 초단 벡터 양자화 복호 신호 계열(121)로서 출력하는 것이다.The third selector 36 outputs the interpolated output vector 125 as the ultra-short vector quantized decoded signal sequence 121 when the index 122 is not a special code, and when the ultra-short index 122 is special protection. Output vector

124 is output as the ultra-short vector quantized decoded signal sequence 121.

FIG. 5A shows the relationship between the block sizes serving as the processing unit of ultra-short vector quantization, and the block size of the initial stage is n ₁ × n ₂ = 16 × 16. FIG. The block size of the blocking is m ₁ x m ₂ = 4 x 4.

Next, the operation of the ultra-short vector quantization encoding and decoding unit 30 will be described with reference to FIG. The first stage vector quantizer 34 is an input vector obtained by blocking the input image signal sequence 101.

와의 왜곡을 구하여 최소 왜곡을 주는 출력벡터

를 탐색한다. 이 왜곡 Ds를, 예를 들면 다음식으로 정의한다.r = n ₁ × n ₂ (n ₁ , n ₂ is a multiple of m ₁ , m ₂ ) and several output vectors read from third to fifth codebooks 31 to 33

Output vector giving minimum distortion by finding distortion with

Navigate. This distortion Ds is defined, for example, in the following equation.

를 식별하기 위한 초단 인덱스(122)를 출력한다.The selected output vector when the minimum distortion Ds is less than a predetermined ultrashort vector quantization threshold

It outputs the first index 122 for identifying.

In addition, when the minimum distortion Ds is larger than the ultra-short vector quantization threshold, it is input to the average value calculator 35 for each of a plurality of samples in the input vector, that is, for each small block on the two-dimensional image sample array, and the ultra-short average value 123 is obtained. At the same time, a special code is assigned to the first index 122.

The vector which becomes this ultra-short average value 123 is interpolated so that it may become the same dimension number as the said input vector, and it will be stored in the 5th codebook 33 as the interpolated vector 125. FIG.

The fifth codebook 33 stores several interpolated vectors 125 and replaces them with the ones stored in the past in time each time the interpolated vector 125 is input to the fifth codebook 33. Are updated sequentially.

Further, in the fourth codebook 32, several output vectors obtained by cutting out blocks at predetermined positions of the past decoded image signal series stored in the frame memory are stored. This stored content is updated together with the contents of the frame memory.

In the third codebook 31, several output vectors having a predetermined level in a uniform pattern are stored in advance.

(124)를 제3의 셀렉터(36)의 전환에 의해 선택되어서 초단 벡터 양자화 복호화 신호 계열(121)로서 출력된다.As described above, when a special code is assigned to the first index 122 in the average value calculating section 35, an output vector.

124 is selected by switching of the third selector 36 and output as the ultra-short vector quantized decoded signal sequence 121.

When the special code is not assigned to the first index 122, the interpolated vector 125 is selected by the third selector 36 and output as the first vector quantized decoded signal sequence 121.

When the ultra-short index 122 is a special code, the ultra-short index 122 is added as the ultra-short vector quantized coded data 120 by adding the ultra-short average values 123.

In the embodiments of FIGS. 3 and 4, generation of an effective block in vector quantization between frames can be suppressed, thereby reducing the amount of code information generated.

In addition, in the codebook 11 of FIG. 1 and the codebook 33 of FIG. 5, the write address is sequentially counted up by one address each time a predetermined vector is inputted. It is also possible to use means for sequentially updating the second codebook 11 and the fifth codebook 33 by controlling the resetting at a time beyond the last address of.

In this way, in the second codebook 11 and the fifth codebook 33, the finite output vectors generated according to the oldest input block that is newest in time with respect to the input block to be encoded can always be stored. have.

Example 2

Next, Example 2 of the present invention will be described. In Fig. 6, reference numeral 37 denotes a sixth codebook constructed similarly to the conventional art, reference numeral 38 denotes a seventh codebook which can be written and read from time to time, and reference numeral 39 denotes a selector for selecting encoded data to be sent. Other portions may be the same as the conventional ones.

Next, the operation will be described. The input vector X 101 is converted into the index i (ⅲ) of the quantization representative vector Y ₁ which gives the minimum distortion by performing the same processing as in the prior art using the sixth codebook 37 in the vector quantization encoder 25. The index 111 is input to the selector 39. Since the seventh codebook is cleared at the start of encoding, the quantization representative vector Y ₁ is selected from the sixth codebook. Here, the magnitude comparison between the minimum distortion di and the arbitrarily set distortion threshold value T is performed, and it is divided into the following two processes according to the comparison result. The selection signal 112 for identifying the process is supplied to the selector 39 and transmitted.

Treatment I: when di <T

The selection signal 112 is set to &quot; 0 &quot;.

The selector 39 outputs and transmits the index i as a coded data 113.

Treatment II: when di> T

The selection signal 112 is set to &quot; 1 &quot;.

The input vector X 101 is output as the encoded data 113 through the selector 39 and transmitted, and the input vector X 101 is written to a predetermined address of the second codebook 38.

By repeating the above process, the input vector when the minimum distortion di exceeds the distortion threshold T is sequentially accumulated in the seventh codebook, and codebooks having different attributes from the sixth codebook are automatically generated.

In the case of the above processing I, a 1-bit prefix indicating which of the sixth and seventh codes belongs to the quantization representative vector Y ₁ is added to the head of the index i.

In order to prevent the seventh codebook from overflowing, an address for writing the input vector X 101 is controlled to repeatedly traverse from address 0 to the maximum address.

In this process, a tree search is used to execute a search of the quantization representative vector Y ₁ at high speed. 7 shows an example of the sixth and seventh codebook configurations in the tree search. The sixth codebook has a binary tree structure as in the prior art, and the seventh codebook has a structure of two layers divided into four classes. The upper quantization representative vectors corresponding to the four classes are set in the same manner as the four quantization representative vectors of the second stage of the sixth codebook, and four quantization representative vectors are stored below each class. Therefore, in the seventh codebook, the search for the 16 quantization representative vectors is performed by searching for the trees of two binary stages. The predetermined value corresponding to the class represented by the upper two bits of the index i of the quantization representative vector Y ₁ giving the minimum distortion selected by the input vector X 101 when the processing II is executed in the sixth or seventh codebook. Is written on the address of. The upper two bits of the index are transmitted together with the input vector X 101.

In the second embodiment, the squared distortion with the input vector is used as an average value when determining the quantization representative vector. For example, as shown in FIG. 8, the input vector is separated into an average value and then normalized to an average of 0 and a size of 1. The seventh codebook is applied to the inner product vector quantization for calculating the inner product with the calculated quantization representative vector and searching for the quantization representative vector having the maximum inner product with the input vector having an average of 0, and the same effect as in the above embodiment. Can be easily realized. In this case, the individual components of the input vector given as the separated average value and the maximum dot product are transmitted independently. Note that the selection signal is not transmitted, and the individual to whom a special code has been assigned is transmitted when the processing II is executed.

Example 3

Next, Embodiment 3 of the present invention will be described with reference to the drawings.

9 is a block diagram showing the configuration of a transmitter of a vector quantizer between frames according to Embodiment 3 of the present invention. In the figure, reference numeral 101 denotes an input image signal sequence, 41 denotes a band pass filter which divides a band of the input image signal sequence 101, and 115 denotes a band divider ( 41 is a first frame memory storing the input image signal sequence 115 for each frequency band divided by the spatial frequency bands by 42), and 116 is a data read signal. 117 denotes a frequency band identification signal output from the first frame memory 42, and 118 denotes a predetermined frequency band input read from the first frame memory by the input of the data read signal 116. An image signal sequence, 119 denotes a prediction signal between predetermined frequency band frames, and (1) a subtraction of the predetermined frequency band input image signal sequence 118 and a prediction signal 119 between this predetermined frequency band frame. Subtractor, (126) Subtractor (1) The differential signal between the predetermined frequency band frames, 43 is a dynamic vector quantization encoder to which the differential signal between the predetermined frequency band frames is input, and 127 is a threshold input to the dynamic vector quantization encoder 43. A value 128 denotes predetermined frequency band encoded data output from the dynamic vector quantization encoder 43, 44 denotes a dynamic vector quantized decoder 301 to which the predetermined frequency band encoded data 128 is input, 301 Is a decoding difference signal between predetermined frequency band frames output from the dynamic vector quantization decoder 44, and (3) is a decoding difference signal 301 between the predetermined frequency band frames and a prediction signal between the predetermined frequency band frames ( An adder for adding 119, 129 is a predetermined frequency band decoded image signal sequence output by the adder 3, 45 is inputted with the data lead signal 116, A second frame memory for generating a prediction signal 119 between the predetermined frequency band frames by giving a frame delay to the predetermined frequency band decoded image signal sequence 129, wherein the predetermined frequency band encoded data A variable length encoder to which 128 is input, 6 is a transmission buffer for speed smoothing connected to the variable length encoder 5, 108 is information generation amount data output from the transmission buffer 6, ( (7) inputs the information generation amount data 108 and the frequency band identification signal from the first frame memory 42, outputs the threshold 127 to the dynamic vector quantization decoder 43, An encoding control unit for outputting the data read signal 116 to the second frame memories 42 and 45, 46 is a line interface (I / F) connected to the transmission buffer 6, and 110 is a line This is a transmission signal output from the I / F 46.

10 is a block diagram showing an example of the configuration of the band dividing section 41, and is composed of M band pass filters into which the same input image signal sequence 101 is input.

11 is a block diagram showing an example of the configuration of the dynamic vector quantization decoder 43. As shown in FIG. In the figure, reference numeral 48 denotes a first block counter which generates the frequency band identification signal 117 according to the difference signal 126 between predetermined frequency band frames in the subtractor 1, wherein An average value separator into which the difference signal 126 between the predetermined frequency band frames and the frequency band identification signal 117 are input, 202 is an average value separation input vector output by the average value separator 8, and 130 is a mean value separator similarly. Average value coded data outputted by (8), (10) stores a normalized output vector, and a codebook into which the frequency band identification signal 117 is input, (211) a normalized output vector outputted by the codebook 10, (9) Is a dot product vector quantization encoder into which the normalized output vector 211, the average value separated input vector 202, the frequency band identification signal 117, and the distortion threshold 72 from the coding controller 7 are input, ( 209) Inner Vector Quantization Part The index of the normalized output vector outputted by the talker 9, 131 is the amplitude coded data output from the dot product vector quantization encoder 9, and 12 is a frequency band identification signal 117 is inputted and the normalized output vector ( The scalar quantizer for the mean value separated input vector 202 passed through the dot product vector quantization encoder 9 without 211 being assigned, 132 is the mean value separated vector coded data output from the scalar quantizer 12, (13). Is a normalization circuit into which the mean value separation vector 202 passed through the dot product vector quantizer 9 is input, and 210 is a normalization input vector output from the normalization circuit 13 to the codebook 10, (28). ) Uses the index 209 in the dot product vector quantizer 9 and the average value separated vector coded data 132 in the scalar quantizer 12 according to the amplitude coded data 131 in the dot product vector quantization encoder 9. Select The fifth selector, 47, is for block identification into which the amplitude coded data 131, the average value coded data 130 at the average value separator 8, and the threshold value 127 at the coded control section 7 are input. And 133 are block identification signals output from the block identification unit 47, and predetermined frequency band coded data 128 output from the dynamic vector quantization coding unit 43 is the block identification signal 133, an average value. The coded data 130, the amplitude coded data 131, and the output of the selector 28 are formed.

FIG. 12 is a block diagram showing an example of the configuration of the dynamic vector quantization decoder 44. As shown in FIG. In the figure, reference numeral 49 denotes a second block counter for inputting the block identification signal 133 in the predetermined frequency band encoded data 128 to output the frequency band identification signal 117, and reference numeral 17 denotes the frequency band identification. An average value decoder to which the average value coded data 130 of the signal 117 and the predetermined frequency band coded data 128 is input, and 18 is a frequency band identification signal 117 and the predetermined frequency band coded data 128. The amplitude decoder to which the amplitude coded data 131 is input, 208 is the amplitude decoder data output from the amplitude decoder 18, and 51 is the frequency determined by the positive and negative of the amplitude decoder data 208. The sixth selector for dividing the signal sent from the fifth selector 28 as part of the band coded data, 209 is the index of the normalized output vector bisected at the sixth selector 51, and 132 is Similarly the sixth selector 51 The average value separated vector encoded data divided by 2, 10 is a codebook similar to that in the dynamic vector quantization encoder 43, 211 is a normalized output vector outputted by the codebook 10, and 50 is The index decoder 55, to which the normalized output vector 211 and the index 209 in the sixth selector 51 are input, 55 outputs the output of the index decoder 50 and the amplitude decoded data 208. A multiplier to multiply, 22 is a scalar quantization decoder to which the average value separated vector encoded data 132 in the sixth selector 51 and the frequency band identification signal 117 are input, and 54 is a scalar quantization. An adder for adding one of the output of the decoder 22 and the output of the multiplier 55 and the output of the average value decoder 17, 53 is a zero signal generator, and 73 is a zero signal generator ( The zero signal output from (53), (52) is used to encode a predetermined frequency band The output of the adder 54 and the seventh selector 301 for selecting the zero signal 73 according to the block identification signal 133 in the second 128 are output from the seventh selector 52. It is a decoding difference signal between predetermined frequency band frames.

13 is a block diagram showing an example of the configuration of the codebook 10. As shown in FIG. In the figure, reference numeral 58 denotes a plurality of fixed codebooks, 59 denotes a plurality of dynamic codebooks, and 56 denotes an eighth selector for selecting the dynamic codebook 59 by the frequency band identification signal 117. And (57) are the ninth selectors for selecting the fixed codebook 58 and the dynamic codebook 59.

FIG. 14 is a block diagram showing an example of the configuration of the coding control unit 7. As shown in FIG. In the figure, reference numeral 68 denotes a data read control unit for outputting the data read signal 116 for instructing read start of data in accordance with the information generation amount data 108, and 69 denotes the first frame memory 42. A block identification threshold table for determining a block identification threshold 71 according to the frequency band identification signal 117 and the information generation amount data 108 given in FIG. A distortion threshold table for determining a distortion threshold value 72 according to the information generation amount data 108, and the threshold value 127 output from the coding control unit 7 includes the block identification threshold value 71 and the distortion threshold value. It is configured as a value 72.

Next, the operation will be described. In FIG. 9, the input image signal sequence 101 is converted into a plurality of frequency band input image signal sequences 115 divided by spatial frequency bands by the band divider 41, which will be described later in detail. The input image signal sequence 115 for each frequency band is stored in the first frame memory 42, and in accordance with the data read signal 116 provided by the coding control section 7, the first frame memory 42 A predetermined frequency band input image signal sequence 118 is read in time division in a predetermined order. The first frame memory 42 simultaneously outputs a frequency band identification signal 117.

The predetermined frequency band input image signal sequence 118 read from the first frame memory 42 is sent to the subtracter 1 to transmit the predetermined frequency band input image signal sequence in the second frame memory 45. The prediction signal 119 between predetermined frequency band frames corresponding to 118 is subtracted, and converted into a difference signal 126 between predetermined frequency band frames. Since the difference signal 126 between the predetermined frequency band frames has a smaller power than the predetermined frequency band threshold image signal sequence 118, encoding with less coding error is possible. As described in detail below, the dynamic vector quantization encoder 43 switches quantization characteristics so that the difference signal 126 between the predetermined frequency band frames inputted in a predetermined order can be adapted according to the height of the spatial frequency. Encoding is performed. In other words, low precision encoding is performed on the difference signal 126 between predetermined frequency band frames having a high spatial frequency in consideration of human visual characteristics, and the difference signal 126 between predetermined frequency band frames having a low spatial frequency is performed. High precision encoding is performed. In addition, the dynamic vector quantization coding unit 43 determines valid blocks, invalid blocks, and vector quantization and scalar quantization with respect to the threshold value 127 provided by the coding control unit 7. The predetermined frequency band encoded data 128 encoded by the dynamic vector quantization encoder 43 is decoded by the dynamic vector quantization decoder 44 and converted into a decoding difference signal 301 between predetermined frequency band frames.

The adder 3 adds the prediction signal 119 between predetermined frequency band frames output by the second frame memory 45 and the decoding difference signal 301 between the predetermined frequency band frames to decode the predetermined frequency band. The image signal sequence 129 is obtained. The predetermined frequency band decoded image signal sequence 129 is temporarily accumulated in the second frame memory 45 to impart a frame delay, and is read in accordance with the data read signal 116 in the encoding control section 7. It is output as a prediction signal 119 between predetermined frequency band frames. On the other hand, the predetermined frequency band coded data 128 is variable length coded by the variable length coding unit 5, temporarily stored in the transmission buffer 6, and subjected to the speed smoothing process to perform the line I / F 46. It is sent out as the transmission signal 110 via. In the transmission buffer 6, the information generating amount data 108 obtained from the storage amount of the variable length coded data is provided to the coding control unit 7. FIG. The encoding control unit 7 includes a threshold 127 composed of a data read signal, a block identification threshold 71 and a distortion threshold 72 in accordance with the information generation amount data 108 and the frequency band identification signal 117. And a data read signal 116 to the first frame memory 42 and the second frame memory 45, and a threshold value 127 to the dynamic vector quantization encoder 43. Control the amount of generation.

Next, the operation of the band divider 41 will be described with reference to FIG. As shown in the figure, the band dividing section 41 is composed of M bandpass filters of # 1 to #M, each having a different passband, and the input image signal sequence 101 is parallel to these bandpass filters. Is entered. Therefore, in each band pass filter, an M type image signal sequence having different predetermined spatial frequency bands is obtained and output as an input image signal sequence 115 for each frequency band.

The operation of the dynamic vector quantization encoder 43 will be described with reference to FIG. The difference signal 126 between predetermined frequency band frames is vectorized in the average value separator 8, and the average value is separated and output as the average value separation input vector 202, while the separated average value is in accordance with the frequency band identification signal 117. After the quantization characteristics are switched and quantized, they are separately output as the average value encoded data 130. The first block counter 48 generates and outputs the frequency band identification signal 117 by counting the difference signal 126 between the predetermined frequency band frames input in a predetermined order in units of blocks. The dot product vector quantization encoder 9 detects the normalized output vector 211 giving the maximum dot product of the dot product of the normalized output vector 211 stored in the average value separation input vector 202 and the codebook 10, The index i 209 is output to the fifth selector 28. Since the amplitude g is equal to the maximum dot product, it is detected simultaneously with the index i 209. When the mean value separated input vector 202 sets Z = (Z ₁ , Z ₂ ,..., Z _K ) and the normalized output vector 211 giving a maximum dot product as Y ₁ , the above dot product vector quantization processing is an example. For example, the following is described.

It is quantized by switching the quantization characteristic corresponding to the spatial frequency band in accordance with the frequency band identification signal 117 of the amplitude g obtained by the above equation, and the amplitude coded data 131 in the dot product vector quantization encoder 9. Is output as. However, when the distortion D between the input and output vectors is larger than the distortion threshold value 72 input from the encoding control unit 7, the amplitude coded data 131 is output with the sign inverted and the fifth selector 28. Index 209 is not output to the scalar quantization coder 12 and the normalization circuit 13, and the mean value separation input vector 202 at the mean value separator 8 is output as it is. The average separated input vector 202 is quantized by the scalar quantization encoder 12 according to the frequency band identification signal 117, and the generated average separated vector encoding data 132 is output to the fifth selector 28. . The average value separation input vector 202 is normalized by the normalization circuit 13 to generate a normalization vector 210.

This normalization vector 210 is stored in the codebook 10 and used as the normalization output vector 211 as described in detail below. In the fifth selector 28, the index 209 in the dot product vector quantization encoder 9 causes the average value separated vector coded data 132 in the scalar quantization encoder 12 to be encoded by the amplitude coded data 131. ) Is selected. The block identification unit 47 determines a valid block / invalid block from the block identification threshold 71 input from the encoding control unit 7, the average value encoded data 130, and the amplitude encoded data 131, and identifies the block. Generate signal 133. Here, the average value encoded data 130 and the amplitude coded data 131, the index 209, or the average value separated vector coded data 132 need not be transmitted to the invalid block.

Next, the operation of the dynamic vector quantization decoder 44 will be described with reference to FIG. The second block counter 49 counts using the block identification signal 133 of the dynamic vector quantization coding unit 43, and generates the frequency band identification signal 117. FIG. The average value decoder 17, the amplitude gain decoder 18, the index decoder 50, and the scalar quantization decoder 22 perform decoding processing using the frequency band identification signal 117. The sixth selector 51 separates the average value of the index 209 sent from the dynamic vector quantization encoder 43 according to the positive and negative of the amplitude decoded data 208 output from the amplitude gain decoder 18. The vector encoded data 132 is identified. The seventh selector 52 identifies the valid block / invalid block based on the block identification signal 133, and selects the zero signal 73 output from the zero generator 53 in the case of an invalid block.

The codebook 10 will be described with reference to FIG. Codebook is composed of # 1~ # L ₁ of the fixed codebook 58 and # 1~ # L ₂ of the dynamic codebook 59 of the L ₂ L _1. When the normalization vector 210 of the normalization circuit is input, the selector 56 of FIG. 8 selects the appropriate dynamic codebook 59 according to the frequency band identification signal 117. In the selected dynamic codebook 59, the codebook is optimized by storing the normalization vector 210 as a new normalization output vector 211 instead of the normalized output vector 211 having a low frequency of use. These several fixed codebooks 58 and dynamic codebooks 9 are selected by a ninth selector 57 operating according to the frequency band identification signal 117, and output an output vector 211 for vector quantization. do.

The operation of the encoding control unit 7 will be described with reference to FIG. The data read control unit 68 counts the information generation amount data 108 to generate a pulse signal for instructing data read start in the first frame memory 42 and the second frame memory 45. It is output as the read signal 116. Meanwhile, the block identification threshold table 69 and the distortion threshold table 70 have several threshold tables for each spatial frequency band, and a predetermined threshold table is selected by the frequency band identification signal 117. The block identification threshold 71 and the distortion threshold 72 are output as threshold values according to the value of the information generation amount data 108 input as shown in FIG. 15 by the predetermined threshold table. do.

Incidentally, in the above embodiment, the data read signal 116 and the threshold 127 output from the coding control unit 7 are determined in accordance with the accumulation amount of the buffer 6, but the circuits using these output signals The determination may be made depending on the data rate or the format of the input image signal sequence 101.

In the above embodiment, the dynamic vector quantization coding unit 43 has described the case where the distortion threshold value 72 given by the coding control unit 7 is input to the dot product vector quantizer 9. The distortion threshold 72 may be input to the scalar quantizer 12 to control scalar quantization characteristics. In addition, the scalar quantizer in the scalar quantizer 12 may be configured as a vector quantizer.

The same effect is also obtained in combination with the motion compensation method of outputting using the contents of the frame memory in the loop.

Example 4

Next, Example 4 of the present invention will be described.

In Fig. 16, reference numeral 10 denotes a fixed codebook constructed in the same manner as in the prior art, reference numeral 60 denotes a variable codebook which can be written and read at any time, and reference numeral 61 denotes a second vector quantization encoder.

Next, the operation will be described. The input vector X 101 is, in the first vector quantization encoder 25, the index i (111) of the quantization representative vector Y ₁ which gives the minimum distortion through the same processing as in the prior art using the fixed codebook 10. The index i 111 is input to the selector 39.

In FIG. 16, the quantization block size was 4x4.

In FIG. 16, the magnitude comparison between the minimum distortion di and the arbitrarily set distortion threshold Ti is performed, and is divided into the following two processes according to the comparison result. In addition, the selection signal 112 for identifying the process is supplied to the selector 39 and transmitted.

The encoding algorithm is described below with reference to the flowchart of FIG.

The index i 111 is output to the selector 39 while the selection signal 112 is set to &quot; 0 &quot;.

The selection signal 112 is set to &quot; 1 &quot; and the input vector X101 is transmitted to the second vector quantization encoder 61.

Next, in the case of the above processing II, two processes are performed by the second vector quantization encoder 61 in succession, but they are described next. The minimum distortion with the input vector was dj, and the quantization block size of the second vector quantization encoder 61 was 2x2.

The index j 135 is transmitted to the selector 39 while the selection signal 134 is set to &quot; 0 &quot;.

The selection vector 134 is set to &quot; 1 &quot; and the input vector X '136 is transmitted to the selector 39, and the input vector X' 136 is written to the variable codebook 60.

Here, read and write operations of the quantization vector to the variable codebook 60 are performed as shown in FIG.

First, a quantization vector and its index number, which divide the feature amount into X and take the value of each feature amount, are registered in order of output frequency. The input vector is immediately extracted with X feature quantities and a quantization vector whose ones match the values of all X feature quantities is selected. If none of the codebooks matches the quantization vector, use the following operation.

(Iii) different for all X of the characteristic quantities.

Let x (α (갖는)) be the input vector column with the first feature value α (ⅰ) and quantization vector string in the codebook be y _j (α (1)).

Select the quantization vector y _j that minimizes the value of and erase it, and write the input vector x instead.

(Ii) When one or more feature quantities are provided, the above formula (*) is calculated from the set of quantization vectors y in which the feature quantities coincide, and y _j is selected and rewritten in the same manner.

In this manner, read, write, and rewrite of the quantization vectors in the codebook are performed.

On the decoder side, the encoded data 113 output from the selector 39 is input to the selector 39, and either one of the first vector quantization encoder 26 and the second vector quantization decoder 62 is used according to the selection signal. (See FIG. 19), when the first vector quantization decoder 26 is selected, the decoding vector Y ₁ 114 is extracted from the fixed codebook 10 according to the index i 111. And output as an output vector. On the other hand, when the second vector quantization decoder 62 is selected, the decoding vector Y _j 137 is extracted from the variable codebook 60 or the input vector X ¹ 136 remains in accordance with the index j 135. It is output as the decoding vector 137. In the latter case, however, the input vector X ¹ 136 is registered in the variable codebook 60 at the same time.

As the block size, the first vector quantization coder 25 and the first vector quantization coder 26 are 4x4, and the second vector quantization coder 61 and the second vector quantization coder 62 are used. Is composed of four 2x2 blocks. Therefore, when the distortion calculation is performed in the first vector quantization encoder 25, and the minimum distortion di takes a value equal to or larger than the threshold value Ti, the original image of the input vector X 101 (it becomes a block size of 4x4). Is divided into four 2x2 blocks, and the minimum vector distortion is calculated by the second vector quantization encoder 61 for each. Accordingly, the successful matching of the variable codebook 60 among the four blocks results in the index j 135 being the coded output, but if the distortion is not very successful, the input vector (2 × 2 size) X ¹ (136). This is output as it is. At this time, it is necessary to write which block is encoded and which block is not encoded as header information of the selection signal 134.

In the above embodiment, two block sizes of vector quantization coders are connected in series, and the first vector quantization coder and the first vector quantization coder are fixed codebooks, the second vector quantization coder and the second vector quantization. A variable codebook is connected to the decoder, but in the case of FIG. 20 in which the variable codebook is connected to the former and the fixed codebook to the latter, the same effect can be obtained as in FIG.

Example 5

Next, Example 5 of the present invention will be described.

21 is a block diagram showing the configuration of the vector quantizer according to the fifth embodiment of the present invention, where 25 denotes a vector quantization of the input vector 101, the index 111 of the quantization representative vector and the minimum distortion ( A vector quantization encoder which outputs 222, (10) stores a plurality of quantization representative vectors (225) and can be written and read from time to time, and (63) controls processing based on the magnitude of the minimum distortion and the distortion threshold value. And a code encoding control unit 64 for outputting an address control signal 223 for updating each quantization vector in the codebook, a new quantization representative vector 224, and encoded data 225, and (64) are obtained from the transmitted encoded data 225. A decoding control section for decoding and outputting the address control signal 223, the new quantization representative vector 224, and the index 226 of the quantization representative vector to be reproduced. Reference numeral 26 denotes a vector quantization decoder which outputs a quantization representative vector 221 corresponding to the index 226 as a decoded vector 114.

Next, the operation will be described.

The input vector X 101 is processed in the vector quantization encoder 25 using the codebook 120 in the same manner as in the prior art, and the quantization representative which gives the minimum distortion di 222 and the minimum distortion di 222. The index i 111 of the vector Y ₁ 221 is output to the coding control unit 63. The encoding control unit 63 performs a magnitude comparison between the minimum distortion di 222 and the arbitrarily set distortion threshold value T. The following two processes are performed according to the comparison result.

(1) when di <T

The process identification signal ″ 0 ″ and the index i 111 indicating that the contents of the codebook 10 are not updated are output as the encoded data 225 and transmitted. In addition, as shown in FIG. 22, tables representing the careers of the transferred indexes are alternately arranged side by side. In other words, if the indexes are side by side in the state 1 now, the index i transmitted this time is placed at the head of the table (the value of the pointer is 1), and in the state 1, the index whose pointer is smaller than the index i (in the drawing) Index i to index r) shifts backward of the table (state 2). By changing the indexes of the tables side by side, the newer and more probable indexes are placed at the head of the table, while the past indexes are placed at the end of the table.

T일 때(2) di

When T

The end of the table is examined to find the most recently sent index. In the case of State 2 in FIG. On the other hand, when the input vector X 101 becomes a new quantization representative vector 224 and the contents of the codebook 10 are updated, the process identification signal ″ 1 ″ and the new quantization representative vector 224 are encoded data 225. The index z is placed at the head of the table and is changed to state 3 in FIG. The quantization representative vector Yz of the index z is rewritten to the new quantization representative vector 224 by the address control signal 223, and the contents of the codebook 10 are updated.

By repeating the above process, the input vector when the minimum distortion 222 exceeds the distortion threshold value T is rewritten with the quantization representative vector transmitted most recently in the codebook, and the contents of the codebook are updated.

The decoding control unit 64 performs the following two processes by the process identification signal decoded in the transmitted encoded data 225.

(1) When the process identification signal is ″ 0 ″

In encoded data 225, the index i '226 of the quantization representative vector is decoded, and the vector quantization decoder 26 decodes the quantization representative vector 221 corresponding to the index i' 226. This codebook 10 is read and output as a decode vector 114. Similarly to the encoding control section 63, a table indicating the history of the transmitted indexes is changed side by side.

(2) When the process identification signal is ″ 1 ″

A new quantization representative vector 224 is decoded from the encoded data 225 and output as a decoding vector 114, and the quantization representative vector 224 is written to the codebook 10 by an address control signal 223, and the encoding side is encoded. Is updated as

Further, in the above embodiment, an index having a high probability of being transmitted from a table obtained by changing the transmitted indexes in parallel is obtained, and if the code allocation control for assigning a shorter code length is performed for the index, the amount of information required for transmission is increased. It is possible to reduce and to perform higher efficiency encoding.

Example 6

FIG. 23 is a block diagram showing a schematic configuration of a vector quantization coding apparatus according to the sixth embodiment, in which the same parts as in FIG. 22 are assigned the same reference numerals, and detailed description thereof is omitted.

In the sixth embodiment, the transmitter A includes a codebook 10a, a vector quantization encoder 25, and an encoding controller 63, and the receiver B includes a decoding controller 64, a vector quantization decoder 26, and a codebook 10b. Consists of.

Next, operation according to the sixth embodiment will be described.

In the transmitter A, the index control signal 223 is outputted from the encoding control unit 63 so as to be arranged side by side in order of increasing frequency according to the frequency of selection of each quantization representative vector in the codebook 10a. In this code assignment, a shorter code is assigned to a quantized representative vector with a higher frequency.

The decoding control unit 64 performs the following two processes by the process identification signal decoded in the transmitted encoded data 225.

(1) When the process identification signal is ″ 0 ″

The index table of the quantization representative vector reproduced in the encoded data 225 is decoded, and the vector quantization decoder 26 reads the vector quantization representative Ymin 228 corresponding to the index data min from the codebook 10b to decode the decoded vector. Output as (114) and "+1" is added to the selectivity of the index data min (227).

(2) When the process identification signal is ″ 1 ″

The new quantization representative vector 223 in the coded data 225 and the index data of the quantization representative vector in the codebook rewritten with the new quantization representative vector are decoded, and the codebook 10b is updated to make a new quantization representative vector 223. Is output as the decoding vector 114, and the selectivity of this index data is set to &quot; 1 &quot;.

By repeatedly executing the above process, the codebook 10b is updated, and the code allocation control outputted by the decoding controller 26 in accordance with the selection frequency of each quantization representative vector in the codebook 10b, similarly to the encoding control unit 63, is output. The signals 223 are set up side by side in order of selection frequency.

Therefore, the codebooks of the encoding side and the decoding side can be matched.

Example 7

Next, a picture coding transmission apparatus according to Embodiment 7 of the present invention will be described with reference to the drawings.

24 is a block diagram showing the configuration of a transmission section of the picture coding transmission apparatus according to the seventh embodiment of the present invention.

25 shows a detailed block configuration of the vector quantization coder.

As shown in the same figure, this vector quantization encoder 25 is used to determine the transmission information of the previous frame temporarily stored in the decision information 146 outputted from the valid / invalid block identification section 14 and the transmission buffer 6. The average value m of the effective difference vector signal 147a output from the quantization control part 96a and the quantization control part 96a which controls the quantization characteristic according to the information amount 148 is calculated, and the average value of the difference vector signal 147a is separated. An average value separation section 96b for performing the following operations; an inner product vector quantization section 96c for vector quantizing the average value separation vector signal 147b2 output from the average value separation section 96b; and a codebook 96d for storing a plurality of pattern vectors. Formed.

Therefore, when the information 146 is valid, the quantization control unit 96a outputs the difference vector signal 145 as an effective difference vector signal 147a as it is, and if it is invalid, outputs a zero vector as the effective difference vector 147a. do.

The average value separating section 96b separates the effective difference vector signal 147a into an average value according to the calculated average value m. The average value separating unit 96b quantizes the average value m in accordance with the quantization characteristic S1 specified by the quantization control unit 96a, and outputs an average value quantization number 147b1.

The average value separation vector 142b2 output from the average value separation unit 96b is output to the dot product vector quantization unit 96c.

Next, the dot product vector quantizer 96c performs an inner product operation by performing an inner product operation with the mean value separation vector 147b2 using a codebook 96d for storing pattern vectors of several mean value separation normalization vectors arranged in a tree structure. Select a pattern vector with a large value.

Next, the operation of the embodiment will be described.

Since the generation of the difference vector signal 145 is the same procedure as in the related art, the description thereof is omitted.

The difference vector signal 145 input to the quantization control unit 96a is adjusted according to the valid / invalid block unit 146 to adjust the value of each element of the vector.

That is, in the case of a valid block, the value of the difference vector signal 145 is output as it is a valid difference vector signal 147a to the average value separating unit 96b. An effective difference vector signal 147a is output.

Since vector quantization coding is not applied to an invalid block, vector quantization coding of a valid block will be described next.

The quantization control unit 96a receives the buffer accumulation information amount 148, which is the transmission information amount of the previous frame, and periodically quantizes the quantization characteristic signal S1 that defines the quantization step width in units of encoding target blocks according to the buffer accumulation amount 148. Will print

The average value separating unit 96b calculates an average value m of the effective difference vector signal 147a, quantizes the resultant signal according to the quantization characteristic signal S1, and outputs the result as the average value quantization number 147b1. The mean value separation of the effective difference vector is performed accordingly, and output to the dot product vector quantization unit 96c as the mean value separation vector 147b2.

Next, the dot product vector quantizer 96c calculates the dot product of the pattern vector normalized to the mean value separation vector 147b2 and the average value 0 and size 1 in the codebook 96d, and indexes the code vector to the maximum dot product. And an amplitude gain 147c1 given as the maximum inner product value are obtained and quantized according to the quantization characteristic signal S1.

Here, the pattern vector is constructed in a tree structure in order to speed up computation when searching for a code vector (quantization representative vector) that gives the maximum dot product in the vector quantization process, and the upper vector is previously designed to be the lower representative vector.

In each stage, an operation of selecting one of two code vectors having the larger inner product with the mean value separation vector 147b2 from the top to the bottom is sequentially performed to determine the lowest quantized representative vector.

In the case of a pattern vector arranged in a binary tree structure, 0 or 1 is allocated according to the separation direction at each node, and a binary sequence representing a path leading to the lowest quantization representative vector is an index of the quantization representative vector ( 147c2).

The new index 147c2 is obtained by cutting off the lower bits of the index 107c2 in accordance with the magnitude of the amplitude gain 0 obtained.

Therefore, in vector quantization coding, since the precision of vector quantization is also variably controlled in accordance with the magnitude of amplitude gain, the amplitude gain quantization number 147c1 can be regarded as the coding precision.

As described above, the encoded vector signal 147 output from the vector quantization encoder 25 has an average value quantization number 147b1, an amplitude gain quantization number 147c1, an index 147c2, valid / invalid block information 146, and quantization characteristics. It is composed of S1, and is output to the variable length encoder 5, multiplexed with the motion information 144b, output to the transmission buffer 6, and output to the communication line as transmission information.

The operation of the vector quantization decoding unit D is replaced with the operation description according to FIG. 12 of the third embodiment.

Next, in FIG. 26, the adaptive spatial filter will be described by showing the arrangement on the screen of the input signal sample pixels for the two-dimensional spatial filter when the two-dimensional spatial filter is used as the adaptive spatial filter.

Here, the sample values of interest to be smoothed and output are X, A, B, C, D, E, F, G, H, respectively, in order from the left to the right of the reference sample pixel values adjacent to each other on a two-dimensional array with respect to this sample. In this case, the smoothing process is executed according to the following equation.

는 평활화된 착안 화소 샘플값, 즉 필터 출력신호 샘플값이고, a₁, a₂, a₃은 평활화 특성을 제어하기 위한 평활화 특성계수이다. 그리고, 상기 평활화 특성계수 a₁을 크게 할수록 평활작용이 약하게 되고, 작게 할수록 평활 작용이 강하게 된다.only,

Is a smoothed pixel value of interest, that is, a filter output signal sample value, and a ₁ , a ₂ , and a ₃ are smoothing characteristic coefficients for controlling smoothing characteristics. As the smoothing characteristic coefficient a ₁ is increased, the smoothing action is weakened. As the smoothing characteristic coefficient a ₁ is increased, the smoothing action is stronger.

An example of an adaptive control method of the adaptive spatial filter 95 will be described below.

As the adaptive control parameters of the adaptive spatial filter 95, the motion position information 144b, the amplitude gain quantization number 147c1 and the valid / invalid block information 146 indicating the encoding precision of the encoding target block at the time of vector quantization encoding are Each is input to an adaptive spatial filter 95.

In the adaptive spatial filter 95, as shown in FIG. 27, the filter characteristics are adaptively controlled for each encoding target block according to the height of the encoding precision indicated by the amplitude gain quantization number 147c1, so that the pixels in the encoding target block are controlled. The smoothing process is performed as a processing unit.

However, when the motion position information 144b is ″ 0 ″ and the valid / invalid block information 146 is an invalid block in the motion compensation process, it is necessary to avoid the high frequency attenuation of the still area, so The smoothing process is not performed for all the encoding target blocks.

In the above embodiment, the adaptive example of the adaptive spatial filter 95 when the vector quantization coding method is used as the block coding method has been described. However, the adaptive filter 95 is also applied to the case where the change coding method or the like is used. In this case, the same effects as in the above embodiment can be obtained by using the coding length after the change coefficient quantization coding of the encoding target block as the coding precision information as the adaptive control parameter.

In addition, in the above embodiment, the adaptive spatial filter 95 has been described in the case of adapting the inter-frame coded transmission apparatus, but the same effect can be obtained when the adaptive spatial filter 95 is applied to the inter-frame coded transmission apparatus.

28 shows another example of the block configuration of an apparatus in which the adaptive spatial filter 95 of the present invention is applied to an encoding transmission apparatus in a frame as another example of the seventh embodiment of the present invention.

In the present embodiment, the block encoding unit 25 performs block encoding on the valid blocks according to the valid / invalid block information 146 in the frame.

The adder of this embodiment adds the decoded vector signal 150 and the reference block 144a in the frame when the input valid / invalid block information 146 is a valid block, while inputting an invalid reference for the invalid block. The adaptive adder outputs the block 144a as a decoded reproduction signal as it is. Then, the adaptive spatial filter 95 is controlled in accordance with the encoding precision information.

In addition, although the embodiment described the embodiment in which the adaptive spatial filter 95 is used as a filter in a loop, the same effect is obtained even when the adaptive spatial filter 95 is applied to the decoding apparatus on the receiving side.

Example 8

Next, Embodiment 8 of the present invention will be described with reference to the drawings. 29 is a configuration example of a block identification unit for executing the coding control method according to the present invention.

Reference numeral 73 denotes a threshold controller for outputting a threshold Th 74 corresponding to the encoding control parameter 109, 74 denotes a threshold Th output from the threshold controller, 80 denotes a comparison threshold T _0. And parameters that display the thresholds and values of the thresholds T ₁ , T ₂ , T ₃ , and T ₄ for each of the areas 1, 2, 3, and 4 shown in diagonal lines in FIG. Input S4, and the value of the threshold T ₁ , T ₂ , T ₃ , T ₄ and the comparison threshold T for each region used in the current frame by the magnitude relationship between the threshold Th and the comparison threshold T ₀ . An identification control unit that determines and outputs a parameter S2 indicating ₀ and an area. Numeral 82 denotes a memory for storing the comparison threshold value T ₀ , the values of the threshold values T ₁ , T ₂ , T ₃ , and T ₄ for each region, and the parameter S for displaying the region and updating each frame. (84) performs block determination based on the values of the threshold values T ₁ , T ₂ , T ₃ , and T ₄ for each region used in the current frame output from the identification control unit 80 to determine the block identification data ν 204. Block version information to output.

This FIG. 29 corresponds to the block identification part 104 in FIG. 43, and therefore, other parts are the same as FIG.

Next, the operation will be described.

If the block identification threshold is equally lowered in units of frames by the block identification unit when the image is stopped, the number of effective blocks for which the value of the block identification data v becomes 1 increases, and the amount of information generation increases rapidly as shown by the solid line in FIG. Thus, the comparison threshold value T ₀ (this T ₀ represents the threshold value of the previous frame when Th is larger than T ₀ and the desired threshold value otherwise) and the threshold controller so as not to increase the amount of information generation rapidly. Compares the threshold value Th (74) outputted from), and when Th is smaller than T ₀ , the threshold value of the entire frame is not lowered to the desired value at a time like the solid line of FIG. The first frame is the first frame, the next frame is the second frame, the next frame is the third frame, and the next frame is the state that Th is not smaller than T ₀ so as to be the threshold value of the entire frame. In the case of 4 frames, the threshold value is gradually lowered to the first, second, third, and fourth frames so as to become closer to the desired values as in the dotted line in FIG. 31, and when the processing of the fourth frame is finished. At this point, the processing is performed so that the threshold value of the entire frame becomes the desired value.

Next, a description will be given according to FIG. The identification control unit 80 determines the magnitude and relationship between the threshold value Th 74 corresponding to the coding control parameter 109 output from the threshold value control unit 73 and the comparison threshold value T ₀ read from the memory 82. The threshold for comparison stored in the memory 82 while updating the values of the threshold values T ₁ , T ₂ , T ₃ , and T ₄ for each area used by the block determination unit 84 according to the value of the parameter S indicated. The value T ₀ , the value of the parameter S indicating the area is updated, and the threshold values T ₁ , T ₂ , T ₃ , T ₄ for each area are compared with the threshold T ₀ for comparison and the value of the parameter S indicating the area. . The block determination unit 84 performs block determination with threshold values T ₁ , T ₂ , T ₃ , and T ₄ for the respective areas output from the identification control unit 80, and outputs the block identification data ν 204.

Next, FIG. 33 which shows the flow of a process according to the structural example of the block identification part in FIG. 29 is demonstrated.

First, the magnitude comparison between the threshold value Th 74 corresponding to the encoding control parameter 109 output from the threshold value control section 73 and the comparison threshold value T ₀ read from the memory 82 is performed. If in the following the magnitude relationship of Th and Th is not greater than T ₀ T ₀ is taken as the value of the parameter S representing a region read out from the memory 82. The following processing is performed based on the magnitude relationship between Th and T ₀ and the value of S.

(0) Stage 0

When Th is _greater than T ₀ , when Th is equal to T ₀ , and when the value of S is 0, the identification control unit 1 updates all values of T ₀ , T ₁ , T ₂ , T ₃ , and T ₄ to Th. The value of S is output as it is and stored in the memory 82 as it is. The block determining unit 84 blocks the entire frame with a threshold of Th and outputs the block identification data ν 204.

(1) First step

When Th is smaller than T ₀ and the value of S is 0, the identification control unit 1 updates the values of T ₀ and T ₁ to Th and the value of S to 1, and T ₂ , T ₃ , T The value of ₄ is output as it is and stored in the memory 82. The block determination unit 84 determines the area 1 as T ₁ = Th = T ₀ , and the areas 2, 3, and 4 as T ₂ , T ₃ , and T ₄ as in the previous step, and determines the block identification data ν 204. Output

(2) second stage

When the value of S is 1, the identification control unit 80 updates the value of T ₂ to T ₀ and the value of S to 2, and outputs the values of T ₀ , T ₁ , T ₃ , and T ₄ as they are. It stores in the memory 82. The block determining unit 84 determines the area 1 + 2 as T ₀ , and the areas 3 and 4 as T ₃ and T ₄ as in the previous step, and outputs the block identification data ν 204.

(3) Third step

When the value of S is 2, the identification control unit 80 updates the value of T ₃ to T ₀ and the value of S to 3, and outputs the values of T ₀ , T ₁ , T ₂ , and T ₄ as they are. It stores in the memory 82. The block determining unit 84 determines the area 1 + 2 + 3 as T ₀ and the area 4 as T ₄ as in the previous step, and outputs the block identification data ν 204.

(4) 4th step

When the value of S is 3, the identification control unit 80 updates the value of T ₄ to T ₀ and the value of S to 0, and outputs the values of T ₀ , T ₁ , T ₂ , and T ₃ as they are. It stores in the memory 82.

The block determination unit 84 blocks the entire frame to a threshold value of T ₀ , and outputs the block identification data ν 204.

In the first, second, third, and fourth steps, if the threshold value Th in the frame in which Th <T ₀ is first observed is set to T, in area (1), area 1 + 2, In (3), the area 1 + 2 + 3 and (4) lower the threshold value of the entire frame to T.

In this embodiment, the area is set to be enlarged radially, and the time until the threshold value of the entire frame is lowered to the desired value is 4 frames. However, this is only one example. The time until the value falls to the desired value is also completely arbitrary.

For example, it is also possible to set the area to be the entire frame at random m cycles (m is a natural number).

In the inter-frame encoding apparatus of FIG. 43, the frame is sequentially encoded in a predetermined order in a predetermined sequence synchronized with transmission and reception, and the region is encoded such that the entire frame is encoded in the frame after a predetermined frame time has elapsed. According to the embodiment, the present invention can also be applied to a mode in which sequence information indicating the frame time elapses is transmitted and an area other than the specified area is encoded between the frames.

Example 9

Next, Example 9 of the present invention will be described with reference to the drawings. 35. The method of claim 34 is also 311 camera, 312 monitor, 313, the camera 311, the ℓ ₁ × 8kbps connection with the monitor (312) to _{(ℓ 1 × 8 + 6.4)} kbps Coding High efficiency encoding device (video codec) of moving picture with speed, 314 is digital data output of this video codec 313, 315 is microphone, 316 is speaker, 317 is l ₂ A speech high efficiency encoding device (voice codec) having a coding rate of 8 kbps, 318 is digital data output of this voice codec 317, and 319 is digital data output 314 at a rate of 1 x 8 kbps. ) And 318 are multiplexed and transmitted to the transmission path 320, 320 is a (digital) transmission path having a capacity of 1x8 kbps connected to the transmission control unit 319.

FIG. 35 is a diagram illustrating a data transmission frame structure in the transmission control unit 319 of FIG. 34, where 321 is a basic time slot of a bit length having a repetition period of 8 KHz, and 322 is a basic time slot ( 321 is a transmission frame composed of 80 pieces.

FIG. 36 is a diagram for explaining the state of the transmission frame transmitted by the transmission control unit 319 of FIG. 34 on the Lx8kbps line, and 323 is bit string data transmitted to the transmission path.

Next, the operation will be described. 34 to turn the invention the ℓ ₁ × 8kbps to _{(ℓ 1 × 8 + 6.4)} kbps video and 2 × 8kbps as an example of application to a video communication terminal having a multiplex communication function of the voice ℓ _1, ℓ ₂ is an integer of 1 or more, ℓ Channels are allocated to video and audio so that a relationship of = l ₁ + l ₂ +1 is established. Wherein the video signal input from camera 311 is encoded in a video deck, nose 313, and sends out the ℓ ₁ × 8kbps to _{(ℓ 1 × 8 + 6.4)} kbps digital data (314). On the other hand, the voice codec 317 encodes and transmits the voice signal input from the microphone 315 to digital data 318 of l ₂ x 8 kbps. The transmission control unit 319 multiplexes the digital data 314 and 318 to 1x8 kbps (l is an integer of 1 or more) and sends it to the transmission path 320. The receiving side performs the reverse operation to decode the digital data 314 and 318, and then outputs the reproduced video signal and the audio signal from the monitor 312 and the speaker 316, respectively.

35, the transmission frame format in the transmission control unit 319 will be described. The basic timeslot 321 has a length of ℓ bits. As in the conventional example, the basic timeslot repetition period Ts is obtained at a transmission rate of × 8 kbps.

Ts = ℓ × 8kbps / ℓbit = 8Hz

It becomes Therefore, the transmission capacity for each sub channel is also given at 8 kbps, and the transmission frame 322 can be configured based on a basic time slot of 8 KHz (eight octave number in the conventional example) as in the conventional example. Therefore, in the present embodiment, for example, when the same frame configuration as that of the conventional example is applied, the transmission frame 322 can be applied to a line of l × 8 kbps even though the same. At this time, the first sub-channel is used as the service channel, and the configuration of the FAS, BAS, and AC is the same as that shown in FIGS. 45 and 46, but there is no problem. It can be applied to.

As a result the 34 cases assigned to ℓ ₁ bit of the ℓ bits in the video, it is possible to assign the bit rate of ℓ ₁ × 8kbps, again by assigning the AC default video of _{(ℓ 1 × 8 + 6.4)} kbps You can assign a ratio. In addition, by allocating L ₂ bits to speech, a ratio of l ₂ x 8 kbps can be assigned to the speech, and this ratio matches the coding rate of the speech codec as described above.

36 shows the state of the bit string 323, but the multiframe configuration and the frame configuration are the same as in the prior art. If the link rate is 56 kbps, then L = 7, and if the kbp is 32 kbps, then L = 4 can correspond to these ratios.

m으로 되는 정수)이라도 좋고, 이 경우에는 m-ℓ비트의 더미데이타를 프레임(324)중에 적당히 배분해 두면 좋다. 이때 더미데이타의 내용을 ″1″로 하는 것으로 다른 전송로에서의 데이타와 논리합을 취하여 다중화하는 것도 가능하다.In the above embodiment, the case where the failure rate of the transmission path 320 is 1 × 8 kbps has been described, but the interface rate with the line is m × 8 kbps (1).

m), and in this case, m-l bits of dummy data may be appropriately distributed in the frame 324. At this time, the content of dummy data is set to "1", and it is also possible to multiplex by taking a logical OR with data from another transmission path.

FIG. 37 shows the frame configuration at this time, in which a transmission frame 325 is formed by arranging a L-bit frame in the m-bit basic frame 324. FIG.

As described above, according to the first embodiment of the present invention, an internal vector quantization encoder is used to update a codebook by scalar quantizing an input vector according to the waveform distortion obtained in the vector quantization encoding process, and use the contents as an output vector in a subsequent encoding process. Since the codebook is sequentially updated according to the vector distortion or the minimum distortion obtained in the quantization process to obtain the amplitude gain, the differential signal sequence between the encoded data and the decoded frame is obtained. Distortion can be suppressed below a certain value, and the amount of encoded information generated and reproduction quality can be widely adapted by changing the threshold for waveform distortion.

It is also possible to generate and update the codebook depending on the local nature of the input image while encoding.

Further, according to another embodiment of the present invention, an output vector of a block image cut out at a predetermined address of the frame memory by providing an ultra-short vector quantization encoder at an output terminal of the frame memory, an output vector of a uniform pattern, and a previous input image signal sequence Since the input image signal sequence is vector quantized from the output vector of the average pattern for each sample to output the ultra-short vector quantized code signal sequence and ultra-short vector quantized coded data, generation of an effective block in vector quantization between frames can be suppressed. There is an effect of reducing the amount of encoded information generated.

In addition, according to the second embodiment of the present invention, a codebook capable of writing and reading from time to time by transmitting an input vector extracted according to the minimum distortion during vector quantization and sequentially storing the input vector as a new quantization representative vector is conventional. The codebook is used together to determine the quantized data that gives the least distortion by tree search, so that the quantization error can be reduced, and the search operation can be performed at high speed.

Further, according to the third embodiment of the present invention, the input image signal sequence is converted into multiple input image signal sequences for each frequency band divided by spatial frequency bands, and the precision of the encoding process is switched in accordance with the spatial frequency of the input image signal sequence. With this arrangement, it is possible to perform effective encoding in accordance with the human visual characteristics, and to improve the subjective quality of the encoded reproduction image.

Further, according to the fourth embodiment of the present invention, by setting the vector quantization decoder connected to the variable codebook, the value of the distortion threshold Ti of the vector quantization coder connected to the fixed codebook can be set small, whereby the SN ratio is increased. The quality improvement is expected to rise. In addition, since the input vector is arranged so that the structure of the variable codebook can be identified for each feature amount, once the feature amount of the input vector is detected, the number of matches can be reduced compared to the conventional method, and the search can be made faster.

In addition, according to the fifth embodiment of the present invention, the input vector extracted according to the minimum distortion in the vector quantization is transmitted, and the input vector is rewritten as the quantization representative vector transmitted most recently in the codebook as a new quantization representative vector. The codebook can be updated, and the codebook can be updated and written from time to time. The quantization error can be reduced, the codebook suitable for the information right is automatically generated, and the change in the input vector can be flexibly followed, so that the quality is transmitted. This has the effect of improving.

According to the sixth embodiment of the present invention, the input vector is rewritten as a scalar quantization representative vector as a scalar quantization representative vector according to the minimum distortion value during vector quantization, and the codebook is rewritten according to the transmission image. The update reduces the quantization error and improves the quality.

In the code assignment, the higher the selection frequency, the shorter the code length index data is allocated, and the coding efficiency is also improved.

Further, according to the picture coding transmission apparatus according to the seventh embodiment of the present invention, the adaptive space filter is arranged in the picture coding transmission apparatus, and the adaptive space filter is arranged in units of blocks according to coding precision, motion vector information, and valid / invalid block information. Since the smoothing characteristic is controlled to be applied adaptively, the coded noise can be removed without degrading the resolution of the decoded picture, so that an effective decoded picture quality can be obtained.

In the interframe coding apparatus using the coding control method according to the eighth embodiment of the present invention, when the image is stopped in the block identification section, the block identification threshold value is not lowered equally, but the block identification threshold value is changed in units of blocks. The generation can be suppressed, and the amount of information generation can be reduced.

In addition, when the image starts to move in the middle of lowering the threshold value, it is of course not accompanied by a sharp increase in the amount of information generation, so that the following can be easily followed.

Further, according to the ninth embodiment of the present invention, it is possible to cope with a line of l × 8 kbps in the same frame structure, making the device compact and inexpensive, and increasing the degree of freedom of the line. There is an effect such as.

As mentioned above, although the invention made by this inventor was demonstrated concretely according to the said Example, this invention is not limited to the said Example, Of course, various changes are possible in the range which does not deviate from the summary.

Claims (17)

The difference picture sequence 103 between frames by subtracting the predicted picture signal sequence 102 between frames read from the frame memory from the frame memory 4 and the input image signal sequence 101 storing at least one frame image signal. Subtractor (1) obtaining?) And the waveform distortion calculated in the process of performing vector quantization decoding with the inner product vector quantization encoder (9) obtained by blocking the predictive image signal between the frames. The minimum distortion obtained in the quantization process in which the input vector is scalar quantized or vector quantized according to the waveform distortion, and the codebooks 10 to 11 are sequentially updated to perform subsequent encoding and decoding as output vectors or to perform average value-normalized vector quantization. The codebook is sequentially updated and coded and decoded according to the value obtained by loading the amplitude gain. And a vector quantization encoding and decoding unit for obtaining a differential signal sequence between the decoded frame, an adder (3) for adding the differential signal sequence between the decoded frames to the predictive image signal sequence between the frames, and outputting the decoded image signal sequence to the frame memory. A variable length encoder 5 for variable length encoding the encoded data and outputting the variable length encoded data, and temporarily storing the variable length encoded data to smooth the amount of information generated and outputting at a constant speed as transmission data, and at one frame unit Vector quantization between frames having a transmission buffer 6 for obtaining an amount of information generation of the encoder and an encoding control unit 7 for generating an encoding control parameter for adaptively controlling the operation of the vector quantization encoding and decoding unit according to the amount of information generation in units of one frame. Encoding and decoding device. The method according to claim 1, wherein the input vector obtained by blocking the input image signal sequence by an ultra-short vector quantizer is quantized by a plurality of output vectors cut out on a predetermined address of the frame memory and output from a codebook. If the minimum distortion obtained in the quantization process exceeds the quantization threshold, the content of the codebook is updated by interpolating the first and second average values obtained by the average value calculating unit, and the interpolated first and second average values are output to the subtractor and adder as an ultrashort vector quantization decoding sequence. At the same time, the ultra-short vector quantization encoding and decoding unit 30, which adds the ultra-short vector quantized decoded data to the variable length encoder by adding the ultra-short average value to the ultra-short index for identifying the output vector, provides a frame at the output of the frame memory. Vector quantization department of liver Chemistry and decoding apparatus. In the vector quantizer which encodes a digital signal with high efficiency, it adds the dynamic codebook 38 which can be written and read several times in the middle of the fixed codebook 37 which consists of several quantization representation vectors arranged in the tree structure. When the minimum distortion in the tree search vector quantization is greater than a predetermined value, the input vector is used as a new quantization representative vector and sequentially accumulated in the predetermined dynamic codebook, and the old quantization representative vector accumulated in the past is erased. By forming a codebook having a learning function, and transmitting identification information of the input vector and the predetermined dynamic codebook in order to match the contents of the dynamic codebook on the transmitting and receiving side, and using the formed codebook and the fixed codebook together. Vector using means to perform vector quantization at high speed Now fire. A band dividing unit 41 for converting an input image signal series into input image signal series for each frequency band divided by spatial frequency bands, and storing the input image signal series for each frequency band output from the band dividing unit for one frame Frame frequency is applied to the first frame memory 42 for outputting a predetermined frequency band input image signal sequence by one time frame for one frame and the predetermined frequency band decoded image signal sequence output from the adder 3, thereby providing a predetermined frequency. A second frame memory 45 for generating a prediction signal between band frames, and a predetermined frequency band frame output from the second frame memory in a predetermined frequency band input image signal series output from the first frame memory. A subtractor 1 for subtracting a prediction signal and generating a difference signal between predetermined frequency band frames, and a predetermined output from the subtractor A dynamic vector quantization encoder 43 for generating a predetermined frequency band encoded data by performing dynamic vector quantization encoding on the difference signal between frequency band frames with a high precision of the spatial frequency, and the first frame memory and the second. A coding control unit 7 for controlling the write / lead of the frame memory and a vector quantization characteristic of the dynamic vector quantization coding unit, and converting the predetermined frequency band coding data generated by the dynamic vector quantization coding unit between predetermined frequency band frames. A dynamic vector quantization decoder 44 for converting into a differential signal is added, and a decoding difference signal between predetermined frequency band frames in the dynamic quantization decoder is added to a prediction signal between predetermined frequency band frames in the second frame memory. A given frequency band An adder 3 for generating an image signal sequence and inputting it to the second frame memory, a variable length encoder 5 for variable length encoding the predetermined frequency band encoded data outputted from the dynamic vector quantization encoder, and the variable Vector quantization between frames having a transmission buffer 6 for temporarily storing variable length coded data in the length encoder and a line interface 46 for transmitting the variable length coded data output from the transmission buffer as a transmission signal. group. 5. The method according to claim 4, wherein the dynamic vector quantization coding unit vectorizes the difference signal between the predetermined frequency band frames to separate an average value, and performs a scalar quantization coding on the average value to perform average value separation input vector and average value coding. An average value separator (8) for outputting data, a fixed codebook storing several normalized output vectors of average ″ 0 ″ and a size ″ 1 ″, and a normalized vector obtained by normalizing the average value separated input vector, as normalized output vectors, The inner product of each normalized output vector stored in the fixed codebook and the dynamic codebook of the codebook is calculated for a codebook 10 including a dynamic codebook whose contents are updated from time to time, and an average value separated input vector given by the average value separator. The above definition giving a maximum dot product of ″ 0 ″ or more Amplitude encoding data defined by the maximum dot product when a distortion obtained according to the magnitude of the average value separated input vector and the maximum dot product is smaller than a distortion threshold value sent from the encoding control unit; Outputs an index for identifying the normalized output vector giving the maximum dot product, and if the distortion is greater than the distortion threshold, inverts the sign of the amplitude encoded data and outputs the mean value instead of the index. A dot product vector quantization encoder 9 for outputting the mean value separated input vector in the separation unit as it is, and a scalar quantization coder for generating average value separated vector encoded data by scalar quantizing the average value separated input vector provided by the dot product vector quantization encoder. 12), the amplitude coding data The selector 28 which selects the index and the average value separated vector encoded data by means of a block counter 48 which counts the difference signal between the predetermined frequency band frames in block units and generates a frequency band identification signal; A block identification unit 47 that determines a valid block / invalid block by using the block identification threshold value sent from the block and outputs a block identification signal, normalizes the average value separated input vector in the dot product vector quantization coder, and normalizes the output vector. And a normalization circuit 13 for generating and outputting to the dynamic codebook in the codebook, wherein the block identification signal, average value coded data, amplitude coded data, and the selector for the valid block are the predetermined frequency band coded data. Selected index or mean value separation vector coding A vector quantizer between frames that outputs one side of data and outputs only the block identification signal for the invalid block. In a vector quantizer for encoding digital signals with high efficiency, a fixed codebook 10, which is a quantization representative vector, is added to the front end, and a variable codebook 60 that can be written and read at a time is added to the rear end, and for each codebook. If an encoding and decoder for vector quantization having different block sizes are connected, and the distortion is not largely encoded in the vector quantization encoder 25 of the previous stage, it is inherent to the input image subdivided by the vector quantization encoder 61 in the subsequent stage. A learning vector quantizer which performs vector quantization while reducing the amount of information generated by encoding at a threshold of. The method of claim 6, further, in order to search for a high-speed and high-precision, representative vector in the variable codebook used in the vector quantizer, the feature vector of the representative vector is divided into several and the codebook area for each feature quantity. Learning type quantizer for registering and updating the representative vector in the variable codebook by arranging the representative vectors having the same characteristics in order of appearance frequency. The method according to claim 6, wherein the variable codebook 60 is connected to the vector quantizer 25 at the front end and the fixed codebook 10 is connected to the vector quantizer 61 at the rear end. A learning vector quantizer which performs vector quantization while reducing the amount of information generated. In a vector quantizer, when performing vector quantization using a plurality of quantization representative vectors stored in a codebook that can be written and read from time to time, the indices of the quantization representative vectors selected from the codebook are changed in order of selection, and the minimum at the time of vector quantization. If the distortion is larger than a predetermined value, a codebook having a learning function capable of rewriting and updating the input vector as a quantization representative vector and rewriting and updating the most recently selected quantization representative vector obtained by side-by-side changing the indices is obtained. And the input vector is transmitted so as to match the codebook on the transmitting and receiving side, and vector quantization is performed using the formed codebook. 10. The control of code assignment according to claim 9, wherein the code allocation is assigned to assign a short coding length to a quantization representative vector having a high selection frequency in accordance with the selection frequency of the quantization representative vector obtained by side-by-side indexing in the vector quantizer. Learning vector quantizer. A codebook which reads an image signal, blocks a predetermined number of pixels at an adjacent position on the image, and generates and outputs an input vector, and a codebook which previously stores several quantization representative vectors to which predetermined index data is added as a pattern of the input vector. 10a) and a vector quantization encoder 25 which selects a quantization representative vector closest to the input vector from the codebook and outputs index data of the quantization representative vector as image transmission data. The frequency of selection of the quantization representative vector is measured, and the code assignment control of assigning short code length index data in order from the quantization representative vector of high selection frequency is performed according to the measurement result, and the quantization representative vector selected during vector quantization is input. Compare the minimum distortion value of the vector with the threshold for vector update. A coding control unit 63 for rewriting an input vector with a quantization representative vector having a low selection frequency when the minimum distortion value is greater than a threshold, and transmitting update identification information, index data, and update representative vector, and received update identification information. According to the rewriting of the quantization representative vector of the receiving codebook, the frequency of selection of the quantization representative vector is measured from the received index data, and short index data are sequentially assigned from the quantization representative vector of the high selection frequency according to the measurement result. And a decoding control unit (64) for updating the index data of the quantization representative vector and the quantization representative vector value in the codebook in accordance with the transmission image. A pixel signal is generated by A / D conversion of a motion image signal every frame, the pixels at positions approaching the image are blocked by a predetermined number, and an image vector signal which is a pixel signal group is generated for each block. The motion processor generates a plurality of reference blocks based on the current block position from the decoding playback signal of the previous frame, searches for a reference block closest to the image vector signal, and detects motion position information of the image vector signal. A vector quantization encoder which compresses and encodes the image vector signal according to the searched reference block and motion position information using correlation between frames, and transmits the encoded image vector signal to a communication line one frame at a time. Decoding obtained by decoding the image vector signal encoded by the buffer and the vector quantization encoder A vector quantization decoder which adds a vector signal of the searched reference block to a vector signal to generate a decoded reproduction signal, and includes a motion compensation process, wherein the coded image is temporarily stored in a transmission buffer. An encoding precision control unit for switching the encoding precision of the vector quantization encoding in a predetermined period in accordance with the amount of transmission encoding information of the vector signal, and setting the pixel value close to each pixel value of the decoded reproduction signal as a pixel value input at a predetermined ratio. Adaptive ON / OFF of the smoothing process of the adaptive spatial filter is controlled according to the adaptive spatial filter for smoothing the decoding reproduction signal and the motion position information, and the smoothing degree of the adaptive spatial filter is low when the coding accuracy is low. When is strong and the coding precision is high And the smoothing degree is weakened, and the image information is provided with a smoothing characteristic control unit. A frame-to-frame encoding apparatus that efficiently encodes an input image signal sequence using conditional pixel supplementation on a block basis, wherein the block used in the past in the process of controlling a block identification threshold value for performing the conditional pixel supplementation on a frame basis; When the desired value smaller than the identification threshold is set as the new block identification threshold, the new block identification threshold is set to a value that gradually decreases while gradually expanding an area in the frame to be applied, and finally to the desired value. And means for performing threshold control applied to the entire frame. The area within the frame to which the new block identification threshold is applied is enlarged to the area designated sequentially according to a predetermined sequence synchronized with transmission and reception, and the entire frame after the predetermined frame time has elapsed by encoding in the area within the frame. The region is set in units of frames so as to encode in a frame, and the sequential information indicating the lapse of the frame time is transmitted for each frame, and the region other than the designated region is inter-frame and inter-frame mixed encoding for inter-frame encoding. And a means for periodically executing the mode, and switching between the intra- and inter-frame mixed encoding mode and the inter-frame encoding mode using threshold control. By specifying a basic tie slot of 1/2/4/6/7 bit length each having a repetition period of 8 KHz corresponding to a transmission rate of 8/16/32/48/56 kbps, the basic timeslot is not dependent on the transmission rate. Means for composing 80/160/320/480 / 560-bit transmission frames each having a repetition period of 100 Hz, each of which is equally collected, and forms the multiframe by collecting the same 16 frames regardless of the transmission speed. Means, means for multiplexing multimedia data such as video encoded data and speech encoded data having an integer multiple of 8 kbps or a code transmission rate matching the transport frame or multiframe, for each of the basic timeslots or transmission frames or multiframes, and the basic time A synchronization code for identifying the division of the transmission frame and the multiframe into a predetermined one bit in a slot and the ratio of the multimedia data One to place the bit allocation information indicating the allocated with a communication control means for notifying the receiving-side multimedia data transmission. The method of claim 15, wherein the number of basic timeslots in the transmission frame and the number of transmission frames in the multiframe are set to a constant value without a transmission rate of 8/16/32/48 / 56kbps. And common information of the synchronization code and the bit allocation information, corresponding to all of the transmission rates with the same control means. The transmission path having a transmission rate of 8/16/32/48 / 56kbps, wherein the bit rate for interfacing is fixed at 64kbps, and 8/16 / for each basic timeslot. Means for converting the transmission rate for speed matching to the interface bit rate by adding dummy bits of 7/6/4/2/1 bit length for each of 32/48/56 kbps and the dummy bit at the receiving side. Transmission rate of 32/48 / 56kbps with means for extracting only valid bits by deleting the dummy bits by division of basic time slots obtained from periodicity or by timing signals of 8 KHz synchronized with basic time slots supplied from the interface unit Multimedia data transmission method characterized by using the same transmission line interface regardless of.

Images