CA1339549C - Video encoding transmission apparatus - Google Patents

Abstract

A video encoding transmission apparatus in which the transmission time of one frame of variable length encoding data becomes less than two times of the video frame time at the transmission side. The apparatus includes a transmission controller for discriminating the idle frame from the transmission frame of the variable length encoding data and for outputting the variable length encoding data per one frame to the memory buffers at the receiving side. A video decoding circuit is used to absorb phase jitter of the video frame period independent of the transmission and receiving sides.

Description

1339~
VIDEO ENCODING TRANSMISSION APPARATUS
This is a division of co-pending Canadian Patent Application No. 615,990 which was filed on January 31, 1991, which in turn is a division of Canadian Patent Application Serial No. 502,755 which was filed on February 26, 1986 and issued to Canadian Patent No. 1,289,656 on September 24, lg91.
The present invention relates to an interframe adaptive vector quantization encoding apparatus which performs encoding of video signals at high efficiency using vector quantization, and also to a video encoding transmission apparatus which transmits the encoding signals.
In accordance with one aspect of the invention there is provided a video encoding transmission apparatus wherein transmission time of one frame of variable length encoding data becomes less than two times of the video frame time at transmission side, said apparatus comprising: a controller at transmission side for performing changing control of write of the transmission buffer with respect to encoding time of the video encoding circuit, i.e., time of two video frames; a transmission control member at transmission side for transmitting a transmission frame with header indicating idle state so as to inform the standby state of variable length encoding data in the transmission buffer to the receiving side; a transmission control member at receiving side for discriminating the idle frame from the transmission frame of variable length encoding data and outputting the variable length encoding data per one frame to the third and fourth buffer memories at receiving side; a controller at receiving 13395~9 side for controlling the third and fourth receiving buffer memories so that the video decoding circuit is transmitted in synchronization with starting of the receiving video frame after finishing of one frame of variable length data transmitted in the time of two frames of the transmission video frame; and a video decoding circuit for performing efficient video decoding of one frame of the variable length encoding data outputted from the third and fourth buffer memories in synchronization with the receiving video frame time and repeatedly outputting the reproduction video signal decoded in the preceding video frame regarding the video frame without the receiving variable length encoding data so as to absorb phase jitter of the video frame period independent at transmission and receiving sides.
The present invention taken in conjunction with the invention disclosed in co-pending Canadian Application Serial No. 615,990 which was filed on January 31, 1991, which in turn is a division of Canadian Patent Application Serial No.
502,755 which was filed on February 26, 1986 and issued to Canadian Patent No. 1,289,656 on September 24, 1991, will be described in detail hereinbelow with the aid of the accompanying drawings, in which:
Fig. 1 is a block diagram illustrating constitution of a vector quantization encoder as an embodiment of the invention;
Fig. 2 is a block diagram illustrating constitution of an encoding member of an interframe adaptive vector quantization encoding apparatus in the prior art;

. 133g5~g Fig. 3 is a block diagram illustrating constitution of a decoding mem~er of an interframe adaptive vector quantization encoding apparatus in the prior art;
Fig. 4 is a block diagram illustrating constitution of a vector quantization decoder in the prior art;
Fig. 5 is a block diagram illustrating constitution of a vector quantization encoder in the prior art;
Fig. 6 is a diagram illustrating operation of a vector quantization encoder as an embodiment of the invention; ~ ~
Fig. 7 is a block diagram illustrating constitution of a vector quantization encoder as another embodiment on the invention;
Fig. 8 is a block diagram illustrating constitution of a vector quantization decoder as another embodiment of the invention;
Fig. 9 is a block diagram illustrating constitution of an encoding member of a vector quantizer as another embodiment of the invention;
Fig. 10 is a block diagram illustrating constitution of a vector quantization member in Fig. 9;

Fig. 11 is a block diagram illustrating 133954 constitution of a code table updating member in Fig. 9;
Fig. 12 is a block diagram illustrating constitution of a decoding member of a vector quantizer as another embodiment of the invention;
Fig. 13 is a block diagram illustrating constitution of an encoding member of an adaptive vector quantizer as still another embodiment of the invention;
Fig. 14 is a block diagram illustrating constitution of a decoding member as still another embodiment of the invention;
Fig. 15 is a diagram illustrating principle of correction in an embodiment of the invention;
Fig. 16 is a block diagram illustrating constitution of an encoding member of a color video signal adaptive vector quantizer as another embodiment of the invention;
Fig. 17 is a block diagram illustrating constitution of a decoding member of a color video signal adaptive vector quantizer as another embodiment of the invention;
Fig. 18 is a diagram illustrating matrix conversion in the quantizer;

1339~49 Fig. 19 is a constitution diagram of an encoding member of a color image efficient encoding apparatus as another embodiment of the invention;
~ig. 20 is a constitution diagram of a decoding member of a color image efficient encoding apparatus as another embodi~ent of the invention;
Fig. 21 is a constitution diagram of a video encoding transmission apparatus as another embodment of the invention;
Fig. 22 is a constitution diagram of a transmission/receiving buffer used in the video encoding transmission apparatus;
Fig. 23-is a constitution diagram illustrating an example OL a transmission fra~e with header; and Fig. 24 is a timing chart illustrating operation of a transmission/receiving buffer.
First, principle of the vector quantization ~ll be described briefly. The input signal series of K
in number are brought together into input vector X = {xl, x2, ... XK}. Then, the K-dimensional Euclidean signal space Rk(X~Rk) has the representati~e points of N in number (i.e., output vector) Yi = {Yil~ Yi2~ ~-- Yik}, and set of the representative points is made 133~ ~4g Y = [Yl~ Y2~ ~-- YN]. The vector quantizer determines output vector ~i as hereinafter described from the set of output vectors, and retrieves it. The output vector Yi is in the minimum distance (minimum distortlon) to the input vector, and it follows that:

if , d ( X , Yi ) < d ( X ~ ) for all X ~ Yi wherein, d(X, ~i) represents distance (distortion) between input/output vectors. Then, the input vector X is transmitted or recorded-in index i of the output vector.
At reproduction, set Y of the output vector _i can be determined by clustering using the video signal series being the training model (repetition of the selection of the representative points and the quantization of the training model into each representative point until the total distortion becoming minimum). Furthermore, in order to improve efficiency of the vector quantization and versatility of the output vector set, the vector quantization may be performed at separation of the mean value of vectors and the normalization in amplitude.
An interframe adaptive vector quantization apparatus in the prior art will be described.
Fig. 2 is a block diagram of an interframe -1339~49 adaptive vector quantization apparatus in the prior art illustrating a constitution example of an encoding member thereof. In Fig. 2, numeral l designates an A/D
converter, numeral 2 a raster/block scan conversion S circuit where digital video signals in raster form by the A/D conversion are made a block per every m picture elements x n lines (m, n: integer), numeral 3 a vector quantization encoder which encodes the block data at high efficiency by means of vector quantization, numeral 4 a transmission data buffer where the encoded data at high efficiency is stored and transmitted to the transmission path at constant speed, numeral 5 a movement detection control circuit which controls threshold value of movement detection in the vector quantization encoder corresponding to the data amount stored in the transmission data buffer, numeral 6 a vector quantization decoder which decodes the encoding data supplied from the vector quantization encoder and reproduces the bloc~ data, numeral 7 a variable delay circuit, and numeral 8 a frame memory.
Fig. 3 is a bloc~ diagram of the interframe adaptive vector quantization encoding apparatus in the prior art illustrating a constitution example of a decoding member. In Fig. 3, numeral 9 designates a 1339~49 receiving data buffer where the encoding data supplied from the transmission path is received and stored and then outputted at speed corresponding to the decoding action, numeral lO a bloc~/raster scan conversion circuit where the block data decoded and reproduced is converted into data in raster form, and numeral ll a D/A converter.
Encoding and decoding operation of the apparatus will now be described referring to Fig. 2 and Fig. 3.
Input video signals (lOl) are analog signals which are raster-scanned from the left to the right on the screen and-from the upper side to the lower side.
The analog signals in raster form are converted into digital signals ~102) by the A/D converter l, and then the digital signals in raster form are made a bLock per every m picture elements x n lines (m, n: integer) by the raster/bloc~ conversion circuit l, and further vector data S (103) is obtained from the picture element sample within the bloc~.
Difference signal between the vector data S
(103) and frame vector data P (104) based on the same position block within the frame memory 8 becomes interframe difference vector ~ (105), which is inputted to the vector quantization encoder 3. The vector quantization encoder 3 applies processing of mean value 1~39~49 separation and normalization to the interframe difference vector E (105), and performs movement detection using mean value and amplitude coefficient obtained by the processing and using threshold value (106), and performs vector quantization of only vector of significant block based on the movement. And then the vector quantization encoder 3 encodes the significance/insignificance information, the mean value, the amplitude coefficient and the vector quantization index information each using variable length encoding or the like, and outputs the encoding data-(107). The transmission data buffer 4 stores the encoding data (107) and transmits it to the transmission path at constant speed according to prescribed transmission speed, and estimates data storage amount (108) and supplies it to the movement detection control circuit 5. The movement detection control circuit 5 controls the threshold value (106) for the mbvement detection depending on variation of the data storage amount (108).
The encoding data outputted from the vector quantization encoder 3 is encoded according to reverse processing of the encoding in the vector quantization decoder 6, thereby interframe difference reproduction vector ~ (lOq) is reproduced. The interframe difference reproduction vector ~ (109) obtained in the processing and - 1339~43 the frame vector data P (104) delayed by prescribed time by means of the variable delay circuit 7 are added, thereby reproduction vector data S (110) is restored and the block data at corresponding position of the frame memory is updated.
On the other hand, the encoding data (107) after received and speed-changed in the receiving data buffer 9 is decoded by the vector quantization decoder 6, thereby interframe difference reproduction vector ~ (109) is reproduced. The interframe difference reproduction vector ~ (109) and the frame vector data P (104) passing through the variable delay circuit 7 are added, and reproduction vector data S (110) is restored in similar manner to the processing in the encoding member. The reproduction vector data S (110) is converted into data (111) of raster form by the block/raster scan conversion circuit 10, and the data (111) is converted in D/A
conversion by the D/A converter 11 so as to obtain analog reproduction video signal (112).
Next, constitution and operation of the vector quantization encoder and the vector quantization decoder will be described in detail. Fig. 5 shows a constitution example of a vector quantization encoder in the prior art.
In Fig. 5, numeral 19 designates a mean value separation 1339~49 and normalization circuit, numeral 13 a movement detection circuit, numeral 23 a distortion operation circuit, numeral 20 a minimum distortion detection circuit, numeral 16 a code book ROM, numeral 17 an address counter, and numeral 18 an index latch.
Operation of the vector quantization encoder will be described.
The mean value separation and normalization.
circuit performs operation as hereinaftPr described to interframe difference vector ~ (113) being input signal thereof, and converts it into normalization vector x.
~Assuming that intrablock mean value of ~ = [~1~ ~2' ~ ~k] (k = m x n) bé m and amplitude coefficient thereof be a, it follows that:

1 !~
m = ~ = 1, 2, ~-- , k ) C ~ m)2 ~1/2 Approximate expr~ssion of ~ such that a = - ~ I E j - m I, a = ma~ ~ - min~
k i=1 J

1~395~9 or the like may be used.
Assuming that m ) / a ~ = L~1 , ~2 ,'~'~ , Z~ ~

the mean value m, the amplitude coefficient a, the normalization vector x can be obtained.
The mean value m (114) and the amplitude coefficient a (118) obtained are inputted to the movement detection circu-it 13, and compared with the threshold values To, Tl, thereby the significance/insignificance block discrimination, i.e., the movement detection processing is performed according to following conditio~s ~5 and the block discrimination information v (121) is outputted.

If m < To and a < TL ; v = 0 (insignificant block) ~If m > To or a > Tl ; v = 1 (significant block) The block discrimination information v is transmitted per each block. Only in the case that v is 1, following processing is performed.
The normalization vector x (122) is transmitted 133~i~9 to the distorsion operation circuit 23 and subjected to following vector quantization processing.
Set of a plurality of output vectors ~i (116) (i = 1, 2, ..., N) is produced using clustering method based on statistical property of the normalization vector x, and written in the code book ROM 16. When the normalization vector x (115) is inputted to the distortion operation circuit 23, the address counter 17 performs count-up in sequence of i = 1, 2, ..., N, and reads the output vector ~i (116) corresponding to the address information i in~sequence of _l~ Y2~ N from the code book ROM 16 of the output vector. Distortions d (x, Yi) (117) between the normalization vector _ and N output vectors Yi read in sequence are calculated sequentially in the distortion operation circuit 23. The distortion calculation is executed according to following formula.

d ( ~ ~ ~i) = C ~ (~j-~ij)2~1/2 ~=1, 2, --k) Otherwise, approximate expression such that d ( ~, ~i ) = ~ I x~ - Yi~

d ( ~, Yi ) = ma~

13~4g may be used.
In the minimum distortion detection circuit 20, the minimum distortion among the N distortions estimated by the above calculation is detected, and the output vector address information i in the code book ROM
indicated by the address counter is taken in the index latch 18 and outputted as the output vector index i (120).
The intrabloc~ mean value m (114), the amplitude coefficient (122), the block discrimination information v (121) and the output vector index i (120), all obtained -in~the above process, are converted into codes being suitable as the vector quantization encoding information (107) and then outputted. In this case, if v is 0, codes in other information are not outputted.
A vector quantization decoder shown in Fig. 4 will now be described.
In Fig. 4, numeral 21 designates an amplitude coefficient multiplier, and numeral 22 a mean value adder.
Among the vector quantization encoding information (107) transmitted from the receiving data buffer 9, the bloc~ discrimination information v (121) is first decoded. -If v = 1, i.e., if the block is significant, the decoded output vector index i (120) is taken in the index latch 1~. In the code book ROM 16 I339 ~9 where the same content as that of the code book ROM of the vector quantization encoder is written, the output vector Yi (119) indicated by the index i (120) is read.
The amplitude coefficient a (118) is multiplied to the output vector Yi in the amplitude coefficient multiplier 21 and the mean value m (11~) is added to the product in the mean value adder 22, thereby the interframe difference reproduction vector ~ (109) is decoded. That is, following processing is executed.

_ C 1 , ~2 , , ~k~

~ Yi~ ~ m ( ~ = i, 2, - , ~ ) If ~ = 0, i.e., if the block is insignificant, e is decoded and reproduced assuming that m = 0, a = 0.

~ = [O, 0, ..., O]

Since the interframe adaptive vector quantization encoding apparatus in the prior art is constituted as above described, calculation of a for normalization and calculation of the square sum (~ (a - b) 2) in the vector quantization distortion operation must be performed at many times, thereby the circuit scale of the apparatus increases. When these calculations are e~ecuted using the 1339~g approximate expressions, the picture quality may be deteriorated due to the calculation error.
Also in the prior art, it is difficult to obtain a universal output vector set which can cope with the picture image having significantly different statistical properties, such as picture image including a document and drawings as main components, picture image requiring the delicate gradation expression or picture image using different sensor system, thereby content of the code book ROM increases.
Further, there exists problem of mismatching in the quantization apparatus.
Moreover, in order to establish the synchroniza-tion of the video frame between receiving and transmission~
time deviation of the top end position between the transmission frame synchronous pattern and the video frame (the number of time slots) must be counted at the transmission side and transmitted to the receiving side, and the top end position of the video frame must be detected from such information at the receiving side and clocks necessary for the decoding processing must be reproduced, thereby the apparatus constitution and the transmission control system become complicated. When the transmission speed is slo~, the information amount which 133~5~g can be transmitted in one video frame time becomes small.
Even if the speed is smoothed at the buffer memory of the transmission side, the transmission efficiency of the encoding data becomes bad and the synchronization of the video frame between transmission and receiving cannot be established.

SUMMARY OF THE INVENTION

The invention relates to an interframe adaptive vector quantization encoding apparatus, wherein analog signals raster-scanned from the left to the right on the screen and from the upper side to the lower side are converted into digital signals by an A/D converter, the digital video signals in raster form are made a block per every m picture elements x n lines (m, n: integer) by a raster/block scan conversion circuit, the block data is -encoded at high efficiency by means of vector quantization in a vector quantization encoder, the encoded data at high efficiency is transmitted to the transmission path at constant speed by a transmission data buffer, threshold value of movement detection in the vector quantization encoder is controlled by a movement detection control circuit, and the encoding data supplied from the vector quantization encoder is decoded and the block data is 133~549 reproduced by a vector quantization decoder. The vector quantization encoder estimates inner product between input vector subjected to mean value separation and output vector read from a code book ROM by an inner product operation circuit, and estimates the maximum value of the inner product by a maximum inner product detection circuit and makes it amplitude coefficient.
According to the invention, since intrabloc~
mean value is separated in a mean value separation circuit and then inner product operation is performed in product sum form and the maximum value is detected so as to estimate the optimum output vector and the amplitude coefficient simultaneously, the operation is simplified and the circuit scale in the apparatus decreases. Since the approximate expression need not be used in the inner product operation, the movement detection and the detection of optimum output vector can be performed at higher accuracy and the reproduction picture quality is improved.
In an encoding member of a vector quantizer as another embodiment of the invention, mean value separated by a mean value separation circuit is subjected to differential PCM encoding thereinafter referred to as "DPCM encoding") thereby redundancy is removed. The 1~39~4~

input vector after the mean value separation is not subjected to amplitude normalization but the inner product operation is performed in product sum form, and the maximum value is detected and subjected to the DPCM
encoding thereby the redundancy is removed, and then the encoding is performed.
According to this embodiment, the amplitude normalization circuit is removed in the vector quantiza-tion encoding member, and the scalar quantization encoding processing of the amplitude coefficient can be simplified and the circuit scale be reduced. Further, the vector quantizer to enable the redundancy removing using the DPCM encoding to the mean value component and the amplitude coefficient can be provided.
A vector quantizer as another embodiment of the invention comprises a code table memory which can rewrite the contact dynamically, cos conversion normalization member for converting the input vector in cos conversion and then performing the normalization processing, a vector quantization member including the code table memory for performing vector quantization of the input vector subjected to the cos conversion normalization, a code table updating member where group of the input vectors of cos conversion normalization quantized into 1~3~4~

the same index in the vector quantization member are added and averaged and written as new set of output vectors of cos conversion normalization in the code table memory, and a normalization restoration inverse cos conversion member for performing reverse processing to the cos conversion normalization member.
According to this embodiment, even if images being significantly different in statistical property are inputted, since the circuit is constituted so that a new set of output vectors by the cos conversion normalization are extracted in clustering and the code table in the encoding member and the decoding member is updated and vector quantization is performed in the cos conversion normalization state, clustering of high speed using a relatively small clustering model without increasing the capacity of the code table enables a set of output vectors with little noise and the vector quantization, and the vector quantization of high quality can be effected to various images.
In a further embodiment of the invention, an amplitude correction circuit is provided to correct amplitude component in the minimum distortion, and the corrected amplitude component is encoded.
According to this embodiment, correction is 13395~9 applied to amplitude while the normalization output vector is reproduced using the quantization distortion, thereby the adaptive vector quantizer is obtained without mismatching of the quantizer.
In still another embodiment of the invention, when color video input signals are composed of three components, red, green and blue, matrix conversion of the three components is performed into luminance component and color difference component, and then the color difference component is sampled and at the same time the mean value of tfie luminance component is separated or the mean value of the luminance component and the mean value of the color difference component are separated, thereby the vector of normalization is sub~ected to vector quantization.
According to this embodiment, a color video vector quantizer to reduce redundancy of any color video signal is obtained using set of a small number of output vectors and increasing the hardware at the least degree.
In another embodiment of the invention, when color video input signals are composed of three components, red, green and blue, each reproduction color video signal after encoding is multiplied by prediction coefficient utilizing corre~ation between channels so as -' 1339549 to form color video prediction signal of each channel, thereby the vector quantization noise is smoothed throughout each channel and accumulatlon of noise in the encoding loop is prevented.
According to this embodiment, a color image efficient encoding apparatus to improve the color reproducibility in region of significant color variation can be obtained without deteriorating the encoding efficiency.
In another embodiment of the invention, a controller is~installed to perform matching in the timing of the video frame between the buffer at trans-mission side and the buffer at receiving side.
According to this embodiment, matching of the timing of the video frame is performed between the transmission and receiving sides without transmitting the top end position information of the video frame and the clock information of decoding to the receiving side, thereby the transmission control system and the control of the transmission and receiving buffers for the speed smoothing can be made easy.

1339~43 _ The embodiments of the invention will now be described in detail.
Fig. 2 is a block diagram illustrating a constitution example of an encoding member of an interframe adaptive vector quantization apparatus of the invention. In Fig. 2, numeral 1 designates an A/D
converter, numeral 2 a raster/block scan conversion - circuit where digital video signals in raster form by the A/D conversion are made a bloc~ per every m picture elements x n lines (m, n: integer), numeral 3 a vector quantization encoder which encodes the block data at high efficiency by means of vector quantization, numeral 4 a transmission data buffer where the encoded data at high efficiency is stored and transmitted to the transmission path at constant speed, numeral 5 a movement detection control circuit which controls threshold value of movement detection in the vector quantization encoder corresponding to the data amount stored in the trans-mission data buffer, numeral 6 a vector quantization decoder which decodes the encoding data supplied from the vector quantization encoder and reproduces the block data, numeral 7 a variable delay circuit, and numeral 8 a frame memory.
Fig. 3 is a block diagram illustrating a 133!~5~9 constitution example of a decoding member of the interframe adaptive quantization encoding ayparatus. In Fig. 3, numeral 9 designates a receiving data buffer where the encoding data supplied from the transmission path lS received and stored and then outputted at speed corresponding to the decoding action, numeral 10 a block/raster scan conversion circuit where the block data decoded and reproduced is converted into data in raster form, and numeral ll a D/A converter.
Encoding and decoding operation of the apparatus will now be described referring to Fig. 2 and Fig. 3.
Input video signals (lOl) are analog signals which are raster-scanned from the left to the right on the screen and from the upper side to the lower side.
The analog signals in raster form are converted into digital signals (102) by the A/D converter 1, and then the digital signals in raster form are made a block per every m picture elements x n lines (m, n: integer) by the raster/block conversion circuit 2, and further the picture element sample within the block is arranged in one-dimensional arrangement so as to obtain vector data S (103). Difference signal between the vector data S
(103) and frame vector data P (104) based on the same 1339~43 position block within the frame memory 8 becomes interframe difference vector ~ (105), which is inputted to the vector quantization encoder 3. The vector . quantization encoder 3 applies processing of mean value separation and normalization to the interframe difference vector ~ (105), and performs movement detecting using mean value and amplitude coefficient obtained by the processing and using threshold value (106), and performs vector quantization of only vector of significant block based on the movement. And then the vector quantization .encoder 3 encodes the sig.nificance/insignificance block information, the mean value, the amplitude coefficient and the vector ~uantization index information each using variable length encoding or the like, and outputs the encoding data (107). The transmission data buffer 4 stores the encoding data (107) and transmits it to the transmission path at constant speed according to prescribed transmission speed, and estimates data storage amount (108) and supplies it to the movement detection control circuit S. The movement detection circuit 5 controls the threshold value (106) for the movement detection depending on variation of the data storage amount (108).
The encoding data outputted from the vector 133~49 quantization encoder 3 is decoded according to reverse processing of the encoding in the vector quantization decoder 6, thereby interframe difference reproduction vector ~ (109) lS reproduced. The interframe difference reproduction vector ~ (109) obtained in the processing and the frame vector data P (104) delayed b~ prescribed time by means of the variable delay circuit 7 are added, thereby reproduction vector data S (110) is restored and the block data at corresponding position of the frame memory is updated. Above process is shown in following expressions.

= S-- P
~ represents operation error in vector quantization encoding.~

,~ ,\
S = P ~ ~ .
P = S ~ Z ( z f represents one frame delay.) On the other hand, the encoding data (107) after received and speed-changed in the receiving data buffer 9 is decoded by the vector quantization decoder 6, thereby interframe difference reproduction vector ~ (109) 1339~49 is reproduced. The interframe difference reproduction vector ~ and the frame vector data P (104) passing through the variable delay circuit 7 are added, and reproduction vector data S (110) is restored in similar manner to the processing in the encoding member.
That is, vector operation S = P ~ ~ = S ~ ~
is executed. The reproduction vector data S (110) is converted into data (111) of raster form by the block/
raster scan conversion circuit 10, and the data (111) is converted in D/A-conversion by the D/A converter 11 so as to obtain analog reproduction video signal (112).
Next, constitution and operation of the vector quantization encoder and the vector quantization decoder will be described in detail.
In Fig. 1, numeral 12 designates a mean value separation circuit which estimates intrablock mean value of interframe difference vector and separates the mean value, numeral 14 an inner product operation circuit which estimates inner product between input vector by the mean value separation and output vector read from a code book ROM, numeral lS a maximum inner product detection circuit which detects the optimum output vector by estimating maximum value of inner products between input 1339~49 and output vectors obtained with respect to a plurality of output vectors and determines amplitude coefficient at the vector quantization decoding, numeral 13 a movement detection circuit which performs movement detection using obtained mean value, amplitude coefficient and threshold value and outputs significance/insignificance block discrimination information, numeral 16 a code book ROM, numeral 17 an address counter, and numeral 18 an index latch.
The operation will be described. The mean value separation circuit 12 performs operation as hereinafter described to interframe difference vector ~ (113) being input vector, and converts it into input vector x*.
Assuming that intrabloc~ mean value cf ~ = C ~ 2, -~
be m, execution of 1 k m = - ~ ~j ( j = 1, 2, -- -, k ) ~*j = ~j- m ~ _ = [~ 1 ~ ~ Z~' - ~ k ~

produces the mean value m and the input vector *.
The inner product operation circuit 14 estimates 133~49 the inner product between the obtained input vector x*
and output vector in the code book ROM 16 where set of output vectors Yi (116)(i = 1, 2, ..., N) optimized by clustering or the like are written using normalization vector subjected to mean value separation and normaliza-y _ [xl, x2, ..., Xk]. The inner product F (x, Yi) is estimated according to following formula.

i=1 Among the obtained inner products of N in number, the maximum value is detected by the maximum inner product detection circuit 15 and at the same time the maximum inner product is outputted as amplitude coefficient ~* to the movement detecting circuit 13.
Assuming that the output vector index to provide the maximum inner product be 0 and the input/out-put vector be two-dimensional vector (k = 2), y0 becomes the optimum output vector and the maximum inner product is obtained as amplitude coefficient. This will be explained referring to Fig. 6.
Since _i is subject to mean value separation and normalization, it follows that I ~i 1= 1 1339 j'49 Assuming that angle between x and Yi be 9, it follows that = I~ I IY i l~~ I '~~ ~

Consequently, when e = o, _ and _i become vectors in the same direction and the distance within two-dimensional plane becomes minimum. Since condition of cos 0 is that -1 < cos 6 < l, when e = o, cos ~ = l and F (_, Yi) becomes maximum.- Hence, yo to maximize the F (x, Yi) is obtained as the optimum output vector.
Since F ( _ , Y~) = l~l IYaI ~~
= ~*IY
= a *

max F (x, Yi) may be used as the amplitude coefficient a*.
The obtained amplitude coefficient ~* and the mean value m are inputted together to the movement detection circuit 13, and compared with the threshold values To~ Tl (106), thereby the significance/insignifi-cance bloc~ discrimination, i.e., the movement detection 133~5~9 processing is performed according to following conditions and the block discrimination information v (121) is outputted.

'If m < To and a < Tl ; v = 0 (insignificant block) If m ~ To or a > Tl ; v = 1 (significant ~ block) The block discrimination information v is transmitted per each block.
The intrablock mean value m (114), the amplitude coefficient (122), the block discrimination information v (121) and the output vector index i (120), all obtained in the above process, are converted into codes being suitable as the vector quantization encoding information (107) and then outputted. In this case, if v = 0, codes in other information are not outputted.
- A vector quantization decoder shown in Fig. 4 will now be described.
In Fig. 4, numeral 21 designates an amplitude coefficient multlplier, and numeral 22 a mean value adder.
Among the vector quantization encoding information (107) transmitted from the receiving data 133~5~

buffer 9, the block discrimination information v is first decoded. If v = 1, i.e., the block is significant, the decoded output vector index i (120) is taken in the index latch 19. In the codè book ROM 20 where the same content as that of the code book ROM of tile vector quantization encoder is written, the output vector _i (119) indicated by the index i (120) is read. The amplitude coefficient a* (118) is multiplied to the output vector _i in the amplitude coefficient multiplier 21 and the mean value m (114) is added to the product in the mean value adder 22, thereby the interframe difference reproduction vector ~ (109) is decoded. That is, following processing is executed.

1~ ~ = C~ J ~2 , , ~
~j = a ~Yii ~ m ( i = 1, 2, - , k ) If v = 0, i.e., if the block is insignificar.t, ~ is decoded and reproduced assuming that m = 0, a = 0.

~ = [~' ~' ' ~]

Fig. 7 and Fig. 8 show another embodiment of the invention. In this em~odiment, an amplitude normalization circuit is removed, and a DPCM encoding 133~5~9 circuit and a DPCM decoding circuit are inserted in vector quantization encoding member and decoding member respectively for mean value component and amplitude coefficient, so that introduction of the DPCM encoding removes redundancy for the mean value component and the amplitude coefficient.
Description of this embodiment is performed regarding the case that threshold control is not performed.
Fig. 7 shows a constitution example of the vector quantization encoding member. In Fig. 7, numeral 12 designates a mean value separation circuit, numeral 14 an inner product operation circuit, numeral 15 a maximum inner product detection circuit, numeral 16 a waveform code book ROM, numeral 17 an address counter, numeral 18 an index latch, numeral 25 a DPCM encoding circuit, and numeral 26 an encoding circuit. Also, Fig. 8 shows a constitution example of the vector quantization decoding member. In Fig. 8, numeral 27 designates a DPCM decoding circuit, numeral 21 an amplitude coefficient multiplier, numeral 22 a mean value adder, and numeral 3 a decoding circult.
In the vector quantization encoding member constituted as above described, the mean value component 1~3~jl9 m (114) is processed in similar manner to the preceeding embodiment, and its redundant component is deleted by the DPCM encoding circuit 25 and the difference signal d~
(123) is outputted.
S On the other hand, the mean value separation vector x* (llS) is processed in similar manner to the preceeding embodiment, and the maximum inner product Pmax (118) is detected by the maximum inner product operation circuit 15, and normalization address informa-tion i is taken in the index latch 18 and outputted as the normalizatian output..vector index i (120). The detected maximum inner product Pmax (118) is inputted as the amplitude coefficient _ directly to the DPCM encoding ~ circuit 25, and its redundant component is deleted and then the difference signal dg (124) is outputted.
The DPCM encoding circuit output d~ (123) for the mean value component m (114), the DPCM encoding circuit dg (124) for the amplitude coefficient g (118), and the normalization output vector index i (120) are.
subjected to variable length encoding in the encoding circuit 26.
Next, the vector quantization decoding operation will be described. The difference signal d~l (123) of the mean value component m, the difference signal dg (124) of 13395g9 the amplitude coefficient g, and the normalization output vector index i (120) are decoded in sequence in the decoding circuit 3.
The mean value m (114) is decoded and reproduced in the DPCM decoding circuit 27, and the amylitude coefficient gi (118) is decoded and reproduced in the DPCM encoding circuit 27. The normalization output vector index i (120) is taken in the index latch 18, and the normalization output vector Yi (116) with address indicated by the normalization output vector index i (120) is read and decQded within the waveform code book ROM 16 where the same content as that of the waveform code book ROM 16 in the vector quantization encoding member is written. The normalization output vector _i (116) is multiplied by the decoded amplitude coefficient gj (123) in the amplitude coefficient multiplier 21, and the decoded mean value component ~ (114) is added to the product thereby the encoding reproduction vector S (109) is obtained.
Still another embodiment of a vector quantization encoding member will now be described.
Fig. 9 shows a constitution example of an encoding member in a video signal conversion vector quantizer. In Fig. 9, numeral 8 designates a video 1339~49 memory where image to be quantized is stored, numeral 28 cos conversion normalization member, numeral 126 cos conversion normalization input vector, numeral 127 d.c.
component and amplitude component of input vector 125, numeral 29 a vector quantization member, numeral 128 vector quantization index, numeral 30 a code table updating member, numeral 129 set of cos conversion normalization output vectors updated, numeral 26 an encoder, and numeral 130 encoding member output signal.
Fig. 10 shows a detailed constitution e.Yample of the vector quantization member 29. In Fig. 10, numeral 31 designates a normalization input vector register, numeral 17 a code table address counter, numeral 130 code table address, and numeral 16 a dynamic cos conversion normalization output vector code table comprising first and second code table memories.
Numeral 32 designates a normalization output vector register, numeral 33 a parallel subtractor for calculat-ing difference between the input and output vectors subjected to cos conversion normalization, numeral 34 a parallel absolute value calculator for outputting absolute value of the difference calculated in the parallel subtractor 33, numeral 23 an absolute value distortion calculation circuit (accumulator), numeral 20 a minimum distortion detection circuit, and numeral 18 an index latch.
Fig. 11 shows a detailed constitution example of the code table updating member 30. In Fig. 11, numeral 35 designates a vector accumulator where the cos conversion normalization input vectors are accumulated per mapped index, numeral 36 an index counter for counting what number of the cos conversion normalization input vectors are mapped on each index, numeral 37 a division calculator where the sum calculated in the vector accumulator 35 is divided by the number counted in the index counter 36, and numeral 38 a register for holding the quotient obtained in the division calculator 37.
Fig. 12 shows a detailed constitution example of the decoding member in video signal conversion vector quantization according to the invention. In Fig. 12, numeral 39 designates a decoder, numeral 132 cos conversion normalization output vector, numeral 40 a normalization restoration inverse cos conversion member, and numeral 133 output vector.
Operation of this embodiment will be described.
Basic principle is in that when images of different properties are inputted and set of output vectors must be 1339~49 updated, clustering is performed using group of input vectors as a model and a new set of output vectors are obtained, the input images are subjected to vector quantization using this set, and the vector quantization information and set of the updated output vectors are made the encoding output so as to cope with significant variation of the input images. In this case, vectors are processed in the cos conversion normalization state. In this constitution, clustering can be converged at high speed using relatively small model, although it is usually converged in a long loop using a large model.
This is because the signal series in the cos conversion is converted from vector in the space axis into vector in the time axis thereby noise component superposed to high frequency component with small power is rapidly removed.
Regarding the signal series f(j) (j = 0, l, ..., M-l), one-dimensional cos conversion F(u) of f(j) becomes as follows:

F(~ f(j)co~ [ 2 ( u = O , 1 , -- , ~-1 ) wherein, C(u)= { l/~ ( u = 0 ) l (u = l,2, 13395~9 Also, inverse cos conversion f(j) of F(u) becomes as follows:

f(j)= ~ C(~ F(~ c~s~ 2 u=~

( i = O, 1, , ~

[Detailed description of the cos conversion is disclosed, for example, in PRATT "DIGITAL IM~GE PROCESSING" (WILEY
INTERSCIENCE).]
Above-mentioned operation will be described referring to the accompanying drawings. In Fig. 9, video signals stored in the video memory 8 are read in form of K-dimensional input vectors (125), S = {Sl, S2, ..., Sk}~
In the cos conversion normalization member 28, for example, as hereinafter described, the cos conversion normalization input vector 126, X = {xl, x2, ..., Xk}, the d.c. component (m) and the amplitude component (a) (127) are formed.

m = (cos conversion first term) = ~ ~ S

~ = ~ j~ 2 I K ~1 S~co~ ~ 2 g ,~

~ ~K ~1 SrCo~ ~ 2 ~ "

( j = 2,3, ,k ) ' Xl = O

That is, the d.c. component m as the first term after the cos conversion is separated from the vector.
The second term and so like are normalized by the mean absolute value sum (amplitude) in each term, and equivalently become the cos conversion normalization input vectors (126) in (K-l) dimensions. The d.c.
component (m) and the amplitude component (a) (127) are separated. Although one-dimensional cos conversion and the mean absolute value sum as amplitude are used in the embodiment, of course, two-dimensional cos conversion or mean square sum may be used.
The cos conversion normalization input vectors X (126) are latched to the normalization input vector register 31 in the vector quantization member 29. The code table address counter 17 sequentially reads the last cos conversion normalization output vectors Yi from the dynamic cos conversion normalization output vector code 13395~9 table 16 by the code table address (130), and latches them to the normalization output vector register 32.
The absolute value distortion operation circuit 23 calculates the absolute value distortion di using the parallel subtractor 33 and the parallel absolute value calculator 34 as follows:

k j _ 1 i 1~

(Term of j = 1 is unnecessary essentially.) The-minimum distortion detection circuit 20 generates strobe signal when the absolute value distor-tion di is minimum, and takes the code table address 130 to the index lutch 18. Content of the index latch 18 at one round of the code table address counter 17 becomes the vector quantization index (120). On the other hand, the cos conversion normalization input vectors (126) are accumulated to the vector accumulator 35 per index (120) quantized in the code updating member 30. The index counter 36 counts the number of vectors mapped to each index. Arithmetic mean of the vectors mapped to each index is calculated by the division calculator 37. The mean vector is held to the register 38, and written as the updated cos conversion normalization output vector 1339~49 Yi (129) in the dynamic cos conversion normalization output vector code table 16. The initial content of the dynamic cos conversion normalization output vector code table 16 is picked up from the cos conversion normaliza-tion input vector X (126).

Yi = ~ (Xi /11 X lli ~Yi~ ~ Yl ,Y2 , ,~N ) (updated set) Wherein, (X)i represents input vector X quantized in the index i, 1IX 11 i represents the number of input vectors X
quantized in the index i, and N represents the number of output vectors. Above-mentioned process is clustering.
When the vector quantization index (128), the d.c.
component and the amplitude component (127), and the code table are updated, the encoder 26 in Fig. 9 performs encoding together with set of the updated cos conversion output vectors (129), and outputs them as the encoding member output signals (125).
In the encoding member as shown in Fig. 12, when the vector quantization index (120), the d.c. component and the amplitude component (127), and the code table are updated in the decoder 39, set of the updated cos conversion normalization output vectors (129) are decoded~

13~9.~9 The updated cos conversion normalization output vectors 129 are written in the dynamic cos conversion normaliza-tion output vector code table 16. The cos conversion normalization output vectors Yi (132) to provide the minimum distortion according to the vector quantization index (120) are latched to the normalization output vector register 32. The normalization restoration inverse cos conversion circuit 40 reproduces the output vectors ( 33), S {S l~ S 2~ --, S'k} using the d.c.
component (m) and the amplitude component (a) as follows:

S j = m + C ' ~ YL~ C0 8 C ( 2 )K( i ) " ~

(j=l ,2, ,k) Another embodi}ltent will now ~e described wherein amplitude component is corrected using minimum quantiz~tion distortion in order to eliminate mismatching of a vector quantizer.
In Fia. 13, numeral 132 designates input vector, numeral 28 a mean value separation normalization circui~, numeral 133 normalization input vec.or, numeral 134 mean value component of the input vector 132~ numeral 135 amplitude component thereof, numeral 31 a normalization 133g.~4~

input vector register, numeral 17 a code table address counter, numeral 131 code table address, numeral 16 a normalization output vector code table memory, numeral 32 a normalization output vector register, numeral 33 a parallel subtractor, numeral 34 a parallel absolute value calculator, numeral 23 an absolute value distortion calculation circuit (accumulator), numeral 20 a minimum distortion normalization output vector detection circuit, numeral 136 minimum quantization distortion detected by t~l~ minimuill distortion normalization output vector detection circuit 20, numeral 137 strobe signal, numeral 18 an index latch, numeral 120 . .
vector ~uantization index, numeral 41 an amplitude correction circuit, numeral 138 corrected amRlitude value, numeral 26 an encoder for encoding the vector quantization index, the mean value component and the amplitude component together, and numeral 130 encoding member output signal.
Fig. 14 shows an example of a decoding member. In Fig. 14, numeral 39 designates a decoder for decoding the vecor quantization index, the mean value component and the amplitude component, numeral 139 normalization output vector, numeral 40 a mean value separation normalization restoration circuit, and numeral 140 output vector.
Operation of this embodiment will be described plur21ity of input signal series are broucilt toaether into 1339~9 input vectors (132), S ={sl, s2, ... , Sk} in block form, and the inpll~ vectors (112) are sep~rated by the mean value separation normal:zation circuit 28 into the mean value component (134) and the amplitude component (135), and the normalization input vectors (L33), X = {xl~ ~2~ k}
are formed. Assuming that the mean va].ue component (13~) be m and the am itude component (135) be a, when the absolute value amplitude is used for example, they are expressed as follows:

3c j - 1 a = K-~ i ~ m¦
j=l ~ (a!~solute value amplitude) xj = 'Sj ~ m)/a (j = 1, 2, ... , k) _ Although the absolute value amplitude is exemplified, k 1 1 a = [X L ~ (sj - m)~ (standar~. deviation) a = 1 ~max(sj - m) - min(sj - m)~

(peak-to-p~a'c v~]ue) or the like, of course, m~y he use~..

S.ince the input vectors aoproach definite 13395~9 distribution within the signal space according to the processing of mean value separation and normalization, ~fficienc~ of the vector quantization is improved. The normalization input vectors (133) are latched to the normalization input vector register 31. The code table address counter 17 sequentially reads t~le normalization output vectors yi from the normalization output vector code table memory 16, and latches them to the normalization output vector register 32. The absolute value distortion calculation circuit 23 estimates the distor-tion di between X and yi from the parallel subtrator 33 and the parallel absolute value calculator 34 as follows:
k di = ~ lxj - Yijl (absolute value distortion) Although the absolute value distortion is exernplifie~, di = ~ (xj _ yij)2 (square distortion) j =l di = m~ax¦xj - Yij¦ (maxirnum element distortion) or the like, of course, may be used.
The minimum distortion detector 20 detects the minimum value of the distortion di between yi and X read in sequence. That is, the minimum distortion becomes d = min di 1~39.~9 ~Ihen the normalization output vector with the minimum distortion is detected, the strobe signal (137) is transmitted to the index latch 18 and the code table address (131) being address of vector is taken and the minimum quanti%ation distortion (136) min di is outputted. At the time of one round of the code table address counter 17, the code table address stored in the inde~ latch 18 becomes the vector quantization index (120). The amplitude correction circuit 41 corrects the amplitude colnponent (135) using t~le minimum quantization distortion (136) according to the strobe signal from the minimum distortion detector 20, and outputs the corrected amplitude value (138). The correction of amplitude is performed in order to minimize the distortion during signal reproduction. ~or example, when quantization distortion is large, the correction corresponds to multiplication of factor to the amplitude so as to decrease the amplitude in the reproduction. In this constitution, not only the distortion due to mismatchinc of the quantizer can be reduced, but also noise component contained in the input signal can be reduced. The amplitude correction circuit 41 may be constituted by ROM having the minimum quantization distortion (136) and the amplitude component (135) as address input, an~ therefore increase of the apparatus scale is quite small. The correction characteristics can be 1~395~9 determined previously in that relation between the quantization distortion min di and the factor multiplied to the amplitude so as to minimize the distortion during reproduction is processed statistically. When the standard deviation is used in the normalization and the square distortion is used as definition of distortion, the correction characteristics may be specified by numerical expression.
Fig. 15 is a diagram illustrating principle of correction in this embodiment using the standard deviation and the square distortion. In Fig. 15, X designates the normalization input vector, and Yi designates the normaliza-tion output vector by vector quantization of X. Since both X and Yi are subjected to mean value separation normalization, they are on hypercircumference of radius ~K in K-dimensional space. Angle ~ between the normalization input vector X and the normalization output vector yi corresponds to distortion of both vectors, and has following relation.
di = 2 K(1 - cos ~) As clearly seen from Fig. 15, in order to obtain the best approximation of X by multiplying factor to yi, co~ ~ becomes the optimum factor. Since above relation is held in similar form also at multiplying the amplitude during reproduction, cos ~ is multiplied as factor to the amplitude, thereby the distortion during reproduction can be reduced corresponding to the quantization distortion di. Since di is estimated during the vector quantization, factor cos ~ is specified as follows:

s di cos 3 = 1 - -Hence, the corrected amplitude value (138) ~' is specified as follows di a' = ~

The encoder 26 encodes the corrected amplitude value (138) ~', the mean value component (134) m and the vector qllantization index 120, and outputs them as the encoding member output signal (130).
In the decoding member, the decoder 39 decodes the vector quantization index (120), the mean value component (134) and the corrected amplitude value (138), and reads the normalization output vectors yi from the normalization out~ut vector code table 16 and latches them to the normalization output vector register 32. Further, the mean value separation normalization restoration circuit 40 decodes the output vectors (140), S' = {51', s2', ... , Sk'}

13~9~4g using the mean value component (134) and the corrected amplitude value (138). That is, Sj' = a ~ Yij + m (j = 1, 2, ... , k) When color video input signals are composed of three components, red, gre~n and blue, the vector quantization including color information is not practicable.
In order to obtain a versatile set of output vectors to enable model formation for any image, the number of the output vectors will become bulky. Consequently, matrix conversion of the three components is performed into luminance component and color difference component, and then the color difference component is sampled and at the same time the mean value of the luminance component is separated or the mean value of the luminance component and the mean value of the color difference component are separated, thereby the vec~or of normalization is subjected to vector quantization. In this constitution, a color video quantizer to reduce redundancy of any color video signal is obtained using set of a small nw~lber of output vectors and increasing the hardware at the least degree.
An embodiment in sucn constitution will now be described in detail. In Fig. 16, numerals 141, 142, 143 designate input signal series of R, G and B components respectively, numeral 42 a matrix conversion circuit, numeral 144 luminance Y signal, numerals 145, 146 color difference I and Q signals, numeral 12 a mean value separation circuit, numeral 147 mean value, numeral 14~
Y signal after mean value separation, numeral 28 a normaliza-tion circuit, numerals 149, 150, 151 Y component, I component and Q component after vector normalization respectively, numeral 152 input vector subjected to luminance mean value separation and normalization, numeral 31 an input vector register, numeral 131 output vector address, numeral 16 an output vector code table, numeral 17 an output vector address counter, numeral 116 normalization output vector, numera.l 117 intervector distortion, numeral 11~ index strobe signal, numeral 153 index signal, numeral 18 an index register, numeral 135 amplitude value, numeral 120 index of minimum distortion output vector, numeral 26 an encoding circuit, and numeral 130 encoding output signal.
Also in Fig. 17, numeral 39 designates a decoding circuit, numeral 154 output vector address, numeral 156 normalization output vector, numeral 21 an amplitude reproduction circuit, numeral 157 mean value separation Yl signal, numeral 158 Il component signal, numeral 159 ~1 component signal, numeral 22 a luminance mean value adding circuit, n~~meral 160 Y component signal, numeral 43 13395~9 a matrix reverse conversion circuit, and numerals 161, 162, 163 R, G, B output signal series.
Operation of this embodiment ~ill be described.
The description will be performed first referring to an example of matrix conversion shown in Fig. 18.
The R, G, B input signal series 141, 142, 143 are converted by the matrix conversion circuit 42 into the Y
component (luminance) and the I, Q components (color difference).
Conversion expressions are as follows:
v = 0.30 R + 0.59 G + 0.11 B
. .
I = 0.60 R - 0.28G - 0.32 B
Q = 0.21 R - 0.52G + 0.31 B
Further, the I, Q components are subsampled.
Fig. 18 shows an example of this conversion. In Fig. 18, the R, G, B signals of 4 x 4 picture elements are converted into the Y signals of 4 x 4 picture elements and the I, O
signals subsampled by the mean values of 2 x 2 picture elements. The luminance Y signals (144) obtained by the matrix conversion eliminate the mean values (147) being d.c. component of luminance in the mean value separ~tion circuit 12, and are transferred as the Y signals (148) of mean value separation to the normali7.ation circuit 28.
Assuming that the Y signal after matrix conversion 13~95~9 be Yj, the mean value separation Y signal be yj and the separated mean value be ~, it follows that Y
16 j=l Yj = Yj - ~

In the normalization circuit 28, the mean value separation Y signals (148) and the color difference I, Q
signals (145), (14G) are brought together (In the e~ample of Fig. 18, Y, I, ~ are brought togetheL into 24-dimensional _ vectors.) and the normalization is performed, thereby the amplitude values (135) of the input vectors are separated.

Assumin~ that the vector before normalization be s, the normalization input vector (152) be u and the amplitude value (135) be a, it follows that s = {yl, ... , yl6, Il, ... , I4, Ql, ~-- , Q4}
= {Sl~ ~-- , 5'4}

a = -24 j=l uj - sj/a U = {ul, ... , U24}
Although the absolute value amplitude is shown as amplitude, the standard deviation, the peak-to-peak value or the like may be usec.

a = { _ ~ s j }Z ; ( standard deviation) 24 j=l a = - {max(sj) - m,n(sj)}
; (peak-to-peak value) The Y, I, Q components (149), (150), (151) after normalization are latched to the input vector register 31, and the input vectors (152) subjectecl to luminallce mean value separation and normalizatiorl are supplied to the distortion calculation circuit. The output vec~or address counter 17 generates the output vector address 131 so that the output vectors (116) are sequentially read from the ou-tput vector code table 16. The distortion calculation circuit 23 calculates the intervector distortion between the normalization input vector _ (152) and the output vector (116) vi from the output vector table 16. Although the absolute value distortion j-l is used in the distortion calculation, the square distortion, the maximum element distortion or the like as hereinafter described may be used.

13395~9 d(u, vi) = ~ (uj - vij) ; (square distortion) The minimum distortion detection circuit 20 supervises the intervector distortion (117) outputted from the distortion calculation circuit 23, and transmits the index strobe signal (119) to the index register 1~3 if the minimum distortion is detected. Then, the irdex signal (153) from the output vector address counter 17 is taken in the index register 1~3 and stored therein. A~ the time of one round of the output vector address counter 17, index within the index register 18 becomes the minimum distortion output vector index (120) and is transmitted to the encoding circuit 26.
The encoding circuit 26 encodes the minimum distortion output vector index tl20) to~ether with the mean value (147) and the amplitude value (135), and outputs the encoding output signal (130) to be transmitted or recorded.
Subsequently in the encoder of Fis. 17, the encoding output signal (130) enters the decoding circuit 39, thereby the output vector index (120), the mean value (147) and the amplitude value (135) are decoded. The minimum distortion output vector index (120) is converted by the index register 18 into the output vector address (154), and the normalization output vectors (156) vl are read from the output vector code table 16.
vi = {Vil~ ... , vi24}
The normalization output vectors (156) are multiplied by the amplitude value (135) in the amplitude reproduction circuit 21, and then the mean value (147) is added to only the mean value separation Y component signal (157) in the mean value adding circuit 22.
s'j = vj x ~
5' = {5'1, .. , 5'.24}
= {Y 1, ~-. , y 16, I 1, .... , I 4, Q 1, , Q~4}
Y' j = ~' j + 1~
The Yl component signals (160) subjected to amplitude reproduction and mean value adding and the I~
component signals (158), (159) after amplitude reproduction are reverse-converted by the matrix reverse conversion circuit 43 into the R, G, B output signal series (161), (162), (163).
~lthough only the Y signal component of the input vector is subjected to the mean value separation in the above embodiment, method of mean valuP separation is not limited to this, but the mean value of the combined component of the I component and the Q component may be separated independently on the mean value of the Y component and subjected to amplitude normalization.
Although the Y, I, ~ signals are obtained from the R, Ç, B signals by the matrix conversion circuit in the above embodiment, the Y, R-Y, B-Y signals may be used in place of the Y, I, ~ signals.
When the color video signals are subjected to the color video difference vector quantization, accumulation of v~ctor quantization noises between the color video sig}lals in the encoding loop may be mutually affected and not averaged, resulting in the significant color shift trailing at the image variation point.
Consequently, each reproduction color video signal after encoding is multiplied by prediction coefficient utilizing correlation between channels so as to form color video prediction signal of each channel, thereby the vector quantization noise is smoothed throughout each channel and accumulation of noise in the encoding loop is prevented. In this constitution, a color image efficient encoding apparatus to improve the color reproducibility in region of significant color variation can be obtained withollt deteriorating the encoding efficiency.
An embodiment in such constitution will now be described. In Fig. 19, numerals 44, 45 and 46 designate 13395~9 subtractors for subtracting color video prediction signals from color video input signals composed of the three components, R, G and B, per each component, and outputting the color video difference signals respectively, numeral 29 a vector quantizer for bringing together the color video difference signals of the three components into vector quantization in the three-dimensional signal space and outputting the color video difference vector quantization signals and the vector quantization inde~, numerals 47, 48 and 49 adders for adding the color video difference vector quantization signals to the color video prediction signals per each component and obtaining the color video reproduction signals respectively, and numerals 50, 51 and 52 R, G and B
prediction devices for forming the color image prediction signals per each component from the color video reproduction signal series respectively.
~umerals 53 through 61 designate multipliers for multiplying the interchannel prediction coefficients of each color video signal respectively, and numerals 62, 63 and 64 accumulators of signals of multiplying the interchannel prediction coefricients to the prediction devices 50, 51 and 5Z of each color video signal per each channel respectively.
Regarding the color video input signals of red, green and blue, assume that the red video input signal (141) 13395~9 be R(D), the green video input signal (142) be G( e~ and the blue video input signal (143) be B(~ herein ~= 1, 2, designates the time series number of each signal. Then, assume that the color video correlation prediction signals (164), (165) and (166) of the red, green and blue channels be P*R(~), P*G(~) and P*g( e, respectively, the color video difference signals (167), (168) and (169) obtained by subtracting the color video correlation prediction signals from the color video input signals throug~ t~ne s~lbtractors 44, 45 and 46 per eacll channel be ~R(~ G(~) and ~B(~) respectively, and the c~lor video vector ~uantization signals (170), (171) and (172) obtained in the vector quantization of the three signals ~R(~!, ~G~i') and ~8(~) by the vector quantizer 29 in the three-dimensional signal space S3 be ~R(~ G(~) and ~B(~). Also assume that the color video reproduction signals (173), (174) and (175) obtained by adding the color video vector signals to the color video correlation prediction signals per each channel be R(Q)~ G(~) and B(~) respectively, and the color video prediction signals (176), (177) and (178) formed in multiplication of prescribed coefficients by the R prediction device 50, the G prediction device 51 and the B prediction device 52 utilizinG correlation of ~he color video signal series per each channel based on the color video reproduction signal series be PR(Q), PG(~) and Pg(~) respectively.
Then, the encodina member shown in Fig. 19 executes the operational processing as follows:
~R(Y) = R(~) - P~ R(~) ~R(~) + QR ( -~ ) ~ G(~) = G(~!) - PG(~) , cG(t) = ~G(t) + QG(~) B ) (~ ) PB(~ ) , ~B( Q) = CB( 1 ) + QB( ~) PR(J~ ) + ~R( ~1) = R(~ ) + QR( ~') G(l) = P~(~) + ~G(P) G( ) QG
PB( ~ ) + ~ B( ~) = B( ~) + QB(~) PR(~) = CR ~ R(~
PG(~) = CG G(~-l) PB(Q ) = CB ~ B( PR(~) = all ~ PR(~) + al2 ~ PG(~) + al3-- P~
PG(J) = a21 ~ PR(~) + a22 ~ PG(j ) + a23 PB( ~ ) PB(l) = a31 ~R(k) + a32 PG(l) + a33 ~ Ps(~) Wherein, all, al2, al3, a21, a22~ a23, a31, a32, a33 are the interchannel prediction coefficients derived from the correlation between the color video signals. The coefficients preferably hold relation as follows:
all + al2 + al3 _ 1 ~ a21 + a22 + a23 - 1 a31 + a32 + a33 < 1 Means for forming the color video correlation prediction signals per each channel by multi~lying the coefficients corresponding to the correlation in the color signals between the adjacent picture elements according to the embodiment, can smooth the vector quantization noises generated by the vector quantizer 29 and prevent the accumulation propagation in the encoding loop. That is, ~[Q~(~)2 + ~G(~)~ +~B(~)]< ~[oR(~)2 + ~C(i)2~.~B(~-)2]

The vector ~uantization is performed in procedure as hereinafter described. The color video difference signals ~R(~)r ~G(i) and ~B(~) in the t~lree-dimensional signal space S3 are brought together into the illpUt vector xe. ~et of the output vectors Yj (j = 1, 2, .... , N) of N in number to map the input vector are prepared, and the output vector to minimize the distance d (x, y~) between the input ~Tector x and the output vector is selected and outputted as the color video vector quantization signals~ R(Q), ~G(~) and~g('~).
The index i of the output vector Yi to provide the minimum distortion is transmitted as the encoding data (130) to the decoding member. Assuming that the vector qllantization processing be VQ, it follows that xe = (xl, x2, x3)~ R(Q), ~G(~)~ ~B(~)) Yi = (Yjl, yj2, yj3) d(x~, y~ (Xk ~ yjk) k=l 133954g V~: XQ -~ Yi if d(~ , yi ~ d(x~, yj) for allj Yi = (yil, yi2, yi3) = (~R(~), 'G(~ B(Y)) Thereby the color video difference vector quantization signals are obtained, and the encoding data is outputted to the decoding member.
Operation of the decoding member as shown in Fig. 20 is performed in that the encoding data (130) of the vector quantizer 29 is received and converted by the vector quantization decoder 39 into the color video difference vector quantization signals (170), (171) and (172) and then the same processing as that of the adding loop in the encoding member shown in Fig. 19 is performed thereby the color video reproduction signals R(t), G(~) and B(~) are obtained. These are improved as shown in the encoding member because the quantization noise is reduced in comparison to the prior art.
Although the above embodiment does not include the case of the intraframe prediction encoding or the adaptive control of the prediction device and the vector quantizer, of course, the invention may be applied to the intraframe or interframe prediction encoding and further to the adaptive changing control of characteristics of the prediction device and the vector ~uantizer.

1339~49 Although the efficient encoding of the color video signals of red, green and blue is described in the em~odiment, the invention can be applied also to the color video signals in combination of the luminance signal and two color difference signals.
Still another embodiment of the invention will be described referring to the accompanying drawings, wherein matching of the timing of the video frame is performed between the transmission and receivill~ sides without transmitting the top end position information of the video frame and the clock inf~rmation of decoding to the receiving side, thereby the transmission control system and the control of the transmission and receiving buffers for the speed smoothing can be made easy.
In Fig. 21, numeral 1 designates an A/D converter where inputted analog video signals are converted into digital data, numeral 65 a synchronous signal generating circuit for generating clocks re~uired for the encoding such as video frame pulse or line pulse from the analog video signals, numeral 66 a video encoding circuit for converting the A/D converter output signal into the encoding data using efficient encoding method, numeral 67 an error correction encoding circuit for applying the error correction encoding to the encoding data transmitted from the ~uffer 13395i9 memory, numeral 68 a transmission path clock source for outputting clocks to transmit the data onto the transmission path at prescribed speed, numeral 69 a frame constitution circuit to constitute the transmission frame for transmitting data together per prescribed time slot in synchronizatioll with the clocks of the transmission path, numeral 70 a frame analysis circuit where data of the transmission frame transmitted through the transmission 2ath are analysed and converted into the encoding data train, numeral 71 an error correction decocing circuit for correcting the error in the reverse processing to that of the error correction encoding circuit, numeral 11 a D/A converter where the digital data decoded by the video decoding circuit is converted into analog video signals and then outputted, numeral 72 a transmission control member at transmission side, and numeral 73 a transmission control member at receiving side.
Numeral 74 designates a header generating circuit for generating a header to discriminate the transmitted data frame, numeral a a transmission buffer where data taken per one video frame and encoded in the video encoding circuit are stored according to comm~nd of a controller and transmitted at prescribed transmission speed, numeral 75 a transmission data multiplication circuit for performing multiplication of the encoding data transmitted from the transmission buffer and other data such as the header, numeral 76 a header interpreting circuit for interpreting the header of the transmitted data frame, numeral 77 a receiving data separation circuit for separating the encodina data and the voice data, numeral 9 a receiving buffer where received encoding data are stored and supplied sequentially to the video decoding circuit 78 according to command of the controller, numeral 79 a decoding synchronous signal generating circuit for generating various cloc~s to perform the decoding action in asynchronous state to the transmission side, and numeral 80 a controller for controlling the buffers at transmission and receiving sides. In Fig. 22, numerals 81, 82 designate buffers 1, 2 having the same capacity, and numerals 83, 84 designate selectors 1, 2 for performing the action changing of the buffers.
Operation of this embodiment will be described.
Digital data outputted from the A/D converter 1 are taken per every other video frame, and then encoded in the video encoding circuit ~ by efficient encoding method with the encoding speed being not constant, such as interframe encoding, and su~plied to the transmission buffer 4. The transmission buffer is a double buffer for performing read/write actions simultaneously, and is constituted as shown in Fig. 22. The buffers 1, 2 are changed in the input/output states by the selectors 1, 2 accordilla to control from the controller 80, and at the same time supply the existing data storage amount to the controller.
The header generating circuit 74 is a circuit to generate header information of the transmission data frame. If the encoding data of the top end position of the video f~ame is transmitted to the transmission fiata multiplication circuit, the header generating circuit 74 outputs the speci~ic data indicating the top end position of the video frame. The header information is multiplied together with the encoding data, the voice data or ,the like in the transmission data multiplication circuit 75. After adding the error correction code, the frame synchronous pattern is inserted in the frame constitution circuit 69, and the arrangement changing of the data train is performed thereby the transmission frame is constituted. A constitution example of tlle transmission frame with header is shown in Fig. 23.
The receiving data train transmitted through the digital transmission 2ath is changed in the arrangement by the frame analysis circuit 70 in synchronization with the frame synchronous pattern in the reverse processing to that as above described. After the error correction in the error correction decoding circuit 71, the header is separated.
The header is discriminated by the header interpreting circui.t 76, and the voice data is separated in the receiving data separation circuit 77 corresponding to constitution of the data frame and the encoding data lS transmitted to the receiving buffer. In the receiving buffer 9, the encoding S data from the data frame indicating the top erld of the video frame discriminated by the one video frame discrimination header to next similar data frame are stored. The encoding data are transmitted to the video decoding circuit 78 in synchronization with the video pulse supplied from the decodlng synchronous signal generating circuit 79. The decoding processing is performed using clocks supplied from the decoding synchronous signal generating circuit, and the data is converted into analog video signals in the D/A
converter 11 and reproduced and outputted.
Relation between operation of the transmission and receiving buffers and the transmission and receiving video frames will be described referring to Fig. 2a. The transmission and receiving buffers take constitution in Fig. 22. At the transmission side, the controller supervises ~Ihether the buffers 1, 2 are empty or not according to the storage amount from the buffers 1, 2, and supplies the control sional of write/read depending on state of each buffer. Assume that the buffer 1 is empty. The encoding data of the frame A is written to the buffer 1 in synchronization with the encoding ON/OFF signal varying at the timing of frame pulse of every other frame. At this time, the buffer 2 reads the frame Y~. If the writing of the buffer l and the reading of the buffer 2 arè finlshed, the buffer l immediztely starts the reading of the frame A and the buffer 2 writes the frame B in s~nchronization with next encoding ON signal.
Subsequently, the reading and writing operation is repeated in similar sequence. Even if the writing of the buffer l is finished (frame C in the figure), while the readinq of the buffer continues (frame B in the figure), the reading of the frame C of the buffer l becomes standby and since the buffer 2 is not empty it cannot write and the frame D is stopped in encoding. On the contrary, even if the reading of the buffer l is finished (frame F in the figure), while the writing of the buffer 2 is not finished (frame G in the figure), until the writing of the buffer 2 is finished, the buffer l reads the dummy data and outputs it as idle transmission frame. According to above procedure, the transmission is performed without stopping the reading operation of the buffers 1, 2.
At the receiving side, the writing operation of the buffers l, 2 is changed alternately and continuously.
If the writing is finished, the buffer immediately performs the reading in synchronization with the frame pulse generated in asynchronous state to the transmission side.
Until the writing of another buffer is finished, the video frame read just before this and decoded is repeatedly outputted in synchronization with the frame pulse thereby the transmission delay amount of the video frame is adjusted.
In this embodiment, the data frame may be constituted in multiplication of the corltrol data by increasing the header information.
In the input signals of the video encoding circuit, every other frame may be taken.
Although the invention can be utilized in a motion picture transmission apparatus for teleconferencing system, it may be applied widely to transmission of motion picture of television without being limited to the teleconferencing system.

Claims

Claims 1. A video encoding transmission apparatus wherein transmission time of one frame of variable length encoding data becomes less than two times of the video frame time at transmission side, said apparatus comprising:
a controller at transmission side for performing changing control of write of the transmission buffer with respect to encoding time of the video encoding circuit;
a transmission control member at transmission side for transmitting a transmission frame with header indicating idle state so as to inform the standby state of variable length encoding data in the transmission buffer to the receiving side;
a transmission control member at receiving side for discriminating an idle frame from the transmission frame of variable length encoding data and outputting the variable length encoding data per one frame to the third and fourth buffer memories at receiving side;
a controller at receiving side for controlling the third and fourth receiving buffer memories so that the video decoding circuit is transmitted in synchronization with starting of the receiving video frame after finishing of one frame of variable length data transmitted in the time of two frames of the transmission video frame; and a video decoding circuit for performing efficient video decoding of one frame of the variable length encoding data outputted from the third and fourth buffer memories in synchronization with the receiving video frame time and repeatedly outputting the reproduction video signal decoded in the preceeding video frame regarding the video frame without the receiving variable length encoding data so as to absorb phase jitter of the video frame period independent at transmission and receiving sides.
2. A video encoding transmission apparatus as set forth in claim 1, wherein a video encoding circuit is provided so that, if the transmission time of one frame of variable length encoding data is quite long in comparison to the transmission video frame time, the efficient video encoding per one video frame at transmission side is controlled at time lapse in the period of two frames.