FR2727545A1 - METHOD AND DEVICE FOR CODING A MOVING IMAGE - Google Patents

Abstract

The present invention relates to a method for encoding a moving image and a device for this purpose comprising first and second buffers (51, 59), a motion estimator / compensator (52), a motion generator / divider (53). motion compensation errors, first and second quaternary shaft processors (54, 55), multiplexer (56), quantizer (57), variable length encoder (58), bit rate controller (60 ), an inverse quantizer (61), a first and second inverse quaternary shaft processors (62, 63), and an adder / filter (64). Motion compensation errors generated after motion compensation are divided into accumulated errors and compensation errors according to their characteristics, and are then encoded. Therefore, the bit output can be reduced compared to the usual method, the image quality can be improved, and the structure of the processing can be simplified.

Description

The present invention relates to a method and a

  device for coding a moving image and, therefore,

  more particularly, a method and a device for coding a moving image in which a motion compensation error (CM error), generated after

  motion compensation, is separated into accumulated errors

  compensation errors in accordance with ca-

  distribution characteristics in a predetermined area

  born, and are coded according to each characteristic

of distribution.

  Generally, a soft image coder

  which has been widely used removes the overlap

  temporal ment through a muscle treatment

  and eliminates spatial overlap by

  median of a discrete cosine transformation (TCD),

to thus compress data.

  Figure 1 shows the basic structure

  tale of a moving picture encoder designed so that temporal overlap can be removed by motion processing and so that spatial overlap can be removed by discrete cosine transformation, and which is intended to be used in a number of standard coders, such as

H.261, MPEG-1 and MPEG-2.

  Referring to Figure 1, an estimator 11

  of motion generates motion vectors for consti-

  kill common images with reference to pre-

  cedents. To generate a motion vector, most

  coders carry out a comprehensive research

  less than a predetermined range based on ab-

  minimum solues (EAM), using a fixed block unit. The motion vector generated by the motion estimator 11 is transmitted to an image memory and a motion compensator 17, and is transformed into an image compensated in terms of motion by compensation.

  of movement in reference to a reconstructed image of ima-

  previous events which are stored in advance in a memory. As a result, the image compensated in terms of

  movement becomes quite similar to the image

  rante due to the motion estimation made by the motion estimator 11 and to the image compensated in terms of motion provided by the image memory and the

motion compensator 17.

  A subtractor 18 calculates the difference between

  the current image and the compensated image in terms of motion

  to generate a motion compensation error

  is lying. So when the image compensated in terms of mo-

  vemnent is added to the motion compensation error

  ment, the complete current image is generated. The mistake

  of motion compensation is transmitted to a dispo-

  sitive of TCD 12, and is then transformed from an area

  in a frequency zone.

  A quantizer (Q) 13 quantifies the motion compensation error, transformed in the device

  of TCD 12, according to a quantification step which is generated

  dré in a device 18 for controlling bit rate.

  The motion compensation error quantified in the

  quantizer 13 is variably coded in length

  in a variable length encoder 19, and trans-

  bet towards a buffer 21, that is to say towards the final stage

  of the encoder, via a multiplexer 20, and is then trans-

put to a decoder.

  In an inverse quantifier (Q-) 14 and in

  a reverse TCD device (TCD-1) 15, serving as a

  reverse transformers of the quantizer 13 and the TCD device 12, the compensation error of

  movement is quantified and transformed by transforma-

  discrete cosine in the opposite direction, respectively-

  ment, and is then added to the image compensated in ter-

  motion measurements in an adder 16, so as to generate a reconstructed image of the current image. The reconstructed image is stored in the image memory and the motion compensator 17. Here, the reconstructed image which is generated is perfectly identical to the reconstructed image generated in the decoder. Consequently, the data transmitted to the

  decoder from the encoder are a compensation error

  motion in which part of the motion vector and data are lost. In the decoder, a motion compensated image is formed using the received motion vector, and then the motion compensation error is inversely transformed, thereby generating a reconstructed image.

  identical to the reconstructed image stored in the memory

  image moiré and motion compensator 17.

  The motion compensation error from

  of the subtractor 18 is generated for two reasons.

  First, when more than two different motion components exist in a block obtained by estimating the

  movement using the fixed block unit via

  diary of the motion estimator 11, the image memory and the motion compensator 17 cannot perform an exact compensation of the motion, this resulting in generating errors from the compensated image in terms of motion and an original image. Here, such an error is called error of

  compensation. Compensation errors have the charac-

  most of which are generated from a boundary of an object moving in

  an image, but few are generated from other em-

  investments. Such an effect is called by Strobach effect of

line drawing.

  Then, the motion compensation error is generated due to an accumulated error. When

  The motion compensation error is fully co-

  dice by one of the following various methods, no error

  is generated from the reconstructed image. However

  However, an error is generated while a quantification

  tion is carried out via a step of quan-

  predification by a device 18 for controlling the bit rate according to a target bit transmission rate. Consequently, errors exist in the reconstructed image generated by the decoder, the memory

  and motion compensator 17. Such er-

  rors are accumulated when treatment continues

  of an image, and do not have a high value but are re-

  games very broadly and randomly.

  Therefore, CM error has the characteristics

  that both a compensation error and an er-

  accumulated error, because the CM error is generated by combining the two errors. That is, the CM error

  has a high value in the limit zone of the ob-

  jet moving in the image like the compensation error

  tion, and a low error value is distributed ma-

  irregular in all areas, the error ac-

  cumulative. The CM error described above is transformed from the spatial area to the frequency area by

  through the TCD device so that a con-

  power centering can be expected. However, a

  power concentration via the dispo-

  sitive TCD is only allowed in the case where the values of each pixel located in the spatial area have a strong correlation between them. Since the CM error has the characteristic of having an irregular distribution due to the accumulated error, when the TCD is performed with respect to the CM error having the irregular distribution, the power is not concentrated, but at

instead is scattered.

  Therefore, Strobach suggested a method of coding through a quaternary tree in

  as a substitute, insisting that the pro-

  TCD cessation, with regard to the CM error, is insufficient

  sucking. In the case of coding using a quaternary tree, when pixels located in blocks have similar values with respect to a block of predetermined size, an average value of the values located in the blocks is coded, but when the pixels located in the blocks have very different values,

  the blocks are separated into four sub-blocks, and the similar

  value litudes in each sub-block are determined

  with regard to the four sub-blocks. Of this ma-

  Finally, an average value of each block and the structure of the quaternary tree are coded by continuously dividing each block into four parts until they

  meet a predetermined condition.

  However, Strobach only considered the characteristic of the compensation errors as a characteristic of the compensation error

  movement, and therefore he paid no attention

  tion to the characteristic of the irregular distribution of accumulated errors. Indeed, the moving image signals have a high dispersion value due not only to errors concentrated on the boundary edge of

  the moving object, but also due to the distribution

  irregular errors accumulated in other zo-

  nes. Therefore, when motion compensation errors are coded through a quaternary tree, all areas must be divided into one

  large number of small parts. As a result, the vo-

  the amount of information increases due to the coding of average values for each zone, comprising information

  on the structure of the quaternary tree to be coded.

  However, since the average value has tended to

  dance to be close to 0 due to the ir-

  regular accumulated errors, a compensatory effect

  tion of remarkable error can only be obtained by in-

intermediate of the average value.

  Figure 2 shows the number of nodes fo-

  Liaries generated when compensation errors, accumulated errors, motion compensation errors (compensation errors + accumulated errors) are divided through a quaternary tree by applying a uniform dispersion threshold 400. Here, we can see that the number of leaf nodes in the compensation errors is much greater than in the accumulated errors. As a result, errors of

  motion compensation increase.

  As mentioned above, the TCD process allows

  tuel widely used for coding error com

  movement thinking has been shown to be ineffective of

  makes irregular distribution of comp errors

  thought of movement and high-frequency components at the edge. In addition, tree treatment

  quaternary proposed as a substitute also occurred

  proven to be ineffective due to irregular distribution

  motion compensation errors.

  However, errors are generated during

  of the quantification step as a function of the quantum step

  tification generated by the bit rate control device 18 due to the data losses generated

  during coding. In the event that the transmission rate

  target binary sion is high, the quantization step is low, and therefore errors are less

  many. However, in the event that the trans-

  target binary mission is weak, the quantification step

  tion is important, and errors are therefore very nom-

  breuses so that the image quality of the images re-

built is deteriorated.

  In particular, in the case of coding a moving image using a very low transmission rate, less than 64 kilo-instructions per second (kcips), the processing of the usual TCD device or

  the quaternary tree structure has a quantification step

  very important due to the low transmission rate

  miter, which results in large data loss. It

  results that phenomena of blocking effect are generated

  cage, in which a persistent image having the same di-

  mension that that of a treated block remains on a screen,

  mosquito effect, in which errors are concentrated

  trés at the limit of a moving object in the form of points, and of discordance of the limit lines of an object, thus resulting in a very good image quality

significantly lower.

  Accordingly, it is a goal of the present in-

  vention to provide a method for coding a moving image in which CM errors generated after

  motion compensation are divided into accumulated errors

  offset and compensation errors according to the ca-

  distribution characteristics in a predetermined area

  born, and are then coded, thereby allowing the

  efficient binaries and reconstructed images

these.

  It is another aim to provide a suitable device

  property for implementing the above process.

  To achieve the above-mentioned object of the present invention, there is provided a method for encoding a moving image comprising the steps of

  (a) generate effective motion vectors

  killing motion estimation for current original images with reference to previous original images, and generating motion compensated images with reference to said motion vectors and

  reconstructed images of the original images previously-

  tes, (b) dividing the motion compensation errors generated after performing said step (a) into accumulated errors and compensation errors,

  (c) treat said er-

  accumulated rors and compensation errors divided during step (b) as a function of the characteristic of each error, (d) quantifying with respect to an average in each quaternary tree said accumulated and compensation errors processed during said step (c), and determining a quantization coefficient as a function of a predetermined quantization step, (e) coding in variable length the data quantified during said step (d), (f) storing the data transmitted from l 'step (e),

  (g) determine the quantization step to be

  to terminate in said step (d) as a function of the volume of said data stored during said step (f), (h) to reverse convert said data quantified in said step (d) into data before quantification,

  (i) carry out a tree treatment four-

  inverse of a block having an average value in a

  predetermined sub-block of the data transmitted from the

  said step (h), to generate compensation errors

  tion and accumulated reproductive errors, and

  (j) generate the first reconstructed images

  tes from said compensation errors and said accumulated reproduction errors generated from said step (i), and said motion compensated images generated from said step (a), and then generate final reconstructed images by

  through low pass filtering.

  To accomplish another purpose of this in-

  vention, there is provided a device for coding a moving image comprising:

  a first buffer intended to memorize images

  previous original ages and original images

  rantes, a motion estimator / compensator

  to generate motion vectors by performing an

  timation of the movement of said original original images

  with reference to said previous original images

  stored in said first buffer, and intended to generate

  dr compensated images in terms of movement in re-

  reference to said generated motion vectors and to the ima-

  reconstructed from said previous original images,

  a generator / divider of compensation errors

  tion of movement intended to generate computation errors

  movement thinking through said ima-

  compensated in terms of movement generated by the-

  said motion estimator / compensator and said ima-

  current original ges, and intended to divide said

  motion compensation errors

  thought and accumulated errors,

  first and second quad shaft processors

  ternary intended for tree treatment

  quaternary so as to code the compensation errors

  tion and the accumulated errors emitted by said genera-

  motion compensation error divider, a switching part for switching

  and to emit said motion vectors generated by the-

  said motion estimator / compensator and the data transmitted by said first and second quaternary tree processors,

  a quantifier intended to quantify the value

  their average in a quaternary tree of the data transmitted

  coming from the switching part, and intended to

  complete a quantization coefficient as a function of a predetermined quantization step,

  a variable length encoder for performing

  killing a variable length coding of said data quantified by said quantizer,

  a second buffer intended to memorize the data

  born from said encoder in length varying

corn,

  a bit rate controller for

  determined to determine the said quantification step to be provided

  audit quantifier as a function of the volume of data

  moralized in said second buffer, an inverse quantifier intended to transform said data quantified by said quantizer into data before quantization,

  first and second quad shaft processors

  reverse ternary intended to perform a reverse quaternary tree processing of blocks with an average in

  a predetermined sub-block of said transmitted data coming from

  from said inverse quantifier, to generate er-

  compensation errors and accumulated repro

  duction, and an adder / filter intended to generate

  first images reconstructed by adding said er-

  compensation rors and said accumulated errors of

  reproduction issued by said first and second process-

  sisters with reverse quaternary tree to said images of com-

  thought of movement emitted by said estima-

  motion compensator, and intended to generate final reconstructed images by handling errors

  irregularly distributed by filtering.

  The objects of the present invention, and others

  advantages of the latter, will become evident

  reading a detailed description of a mode

  preferred embodiment, given as a unique example-

  ment, made with reference to the attached drawings on the-

which:

  - Figure 1 shows the structure of a co-

  deur of images in usual movement, - figure 2 represents the number of leaf nodes generated when the compensation errors, the accumulated errors and the compensation errors of

  movement are treated by a quad shaft treatment

ternary,

  FIG. 3 is a functional diagram represented

  feeling a device for coding a moving image according to the present invention,

  - Figure 4 is a block diagram

  cut from a motion estimator / compensator and an error generator / divider shown in FIG. 3,

  - Figure 5 is a flowchart for it-

  polish the operation of the first and second processes

  sisters with a quaternary tree represented in FIG. 3,

  - Figure 6 is a block diagram

  cut from an adder / filter shown in Figure 3, - Figure 7 shows examples showing

  schematically a related processing structure

* tive to accumulated errors and compensation errors

tion, and

  - Figure 8 shows sequences of experiment

  filtering experience regarding accumulated errors

  both after processing compensation errors.

  We will describe the present invention in detail,

  referring to the accompanying drawings.

  Referring to Figure 3, a device

  to encode a moving image according to the present in-

  vention includes: first and second buffers 51 and

  59, a motion estimator / compensator 52, a general

  motion compensation error generator / divider 53, first and second quaternary tree processors

  54 and 55, a switching part, for example, a multi-

  tiplexer 56, quantifier 57, length coder

  variable 58, a bi-flow control device 60

  nary, an inverse quantizer 61, of the first and se-

  cond processors with reverse quaternary shaft 62 and 63, and

an adder / filter 64.

  When the device for coding the image in

  movement shown in Figure 3 is in operation

  ment, the first buffer 51 stores current and previous original images, the estimator / compensator of

  motion 52 generates motion vectors in effect

  killing the estimation of the movement compared to the images

  with reference to the original images above-

  your stored in the first buffer 51, and generates compensated moving images with reference to the vectors

  generated movements and reconstructed images of ima-

  previous original events which are stored.

  The compensation error generator / divider

  motion 53 generates motion compensation errors using motion compensated images generated by the motion estimator / compensator 52

  and current original images, and generates er-

  accumulated errors and errors of compensation in divi-

  health motion compensation errors.

  The first and second quad shaft processors

  ternary 54 and 55 each perform quaternary tree processing to code the errors of

  compensation and accumulated errors issued by the ge-

  motion compensation error generator / divider 53.

  Multiplexer 56 receives motion vectors

  ment generated by the motion estimator / compensator 52 and the data sent by the first and second pro-

  quaternary tree stoppers 54 and 55, selects ma-

  sequentially the input motion vectors and the

  entered data, and transmits them to a decoder (not

represented).

  The quantizer 57 performs a quantification

  tion compared to an average value during processing

  ment by quaternary tree from the data supplied by the multiplexer 56, and determines a quantization coefficient in accordance with a quantization step

  supplied by the bit rate control device 60.

  The variable length coder 58 performs variable length coding of the data quantized by the quantizer 57, and a second buffer 59 stores the

  variable length encoded data 58.

  The bit rate control device 60

  determines the quantization step to be provided to the quantifier

  encoder 57 as a function of the volume of data stored in the second buffer 59, and the inverse quantizer 61 performs an inverse operation to that of the quantizer 57 and transforms the quantized data in the form of

  data similar to that available before

tification.

  First and second quad shaft processors

  ternary inverse 62 and 63 generate cornm errors

  thought of reproduction and accumulated errors of reproduction by extracting only blocks having

  an average value in a predetermined sub-block, from

  from the data sent from the quantizer

pour 61.

  An adder / filter 64 generates first

  res images reconstructed by adding accumulated errors

  mules and reproduction compensation errors to motion compensated images that are output from the motion estimator / compensator 52, and then generates final reconstructed images by processing

  errors distributed irregularly by wire-

trage.

  To begin with, the characteristics of er-

  motion compensation rors having both the

  of accumulated errors and that of compensation errors, can be expressed as follows: et = MC (ME (xti), xt-i) - xt = le__. + ecel ... (1) o and indicates the motion compensation errors, t indicates the time, ME (k) indicates the motion vectors generated with reference to an image "k", MC (v, k) indicates

  an image reconstructed using moving vectors

  ment "v" and with reference to the image "k", xt indicates a

  reconstructed image for a time period "t" and xt indi-

  as an original image for the time period "t", respectively. In the following formula (2), the motion compensation errors "and" expressed by the formula (1) are divided into compensation errors (eà._c) and in

  accumulated errors (eacc), so that processing

  property of the distribution characteristics of each

error is possible.

  et = MC (ME (xtl), xt-1) - xt = MC (ME (xt-1.), xt) - xt + eacc ... (2) = emc + eacc

  Here, "and" can be obtained by formula (1).

  Since "emc =" is' "C (ME (xt-1), xt-,), the accumulated errors and the compensation errors can be

  obtained by "eac = et - es.". Compensation errors

  tion are distributed focusing on the outer edge

  laughing at the moving object and have a very small value

  in other areas. In the event that the errors of com-

  thinking whose values have already been concentrated in the limited area are divided into small parts by the quaternary tree method, efficient coding is

  possible. In the other zones, the coding is not effective.

  effected by the fact that the mean value is close to 0.

  However, the accumulated errors are very ir-

  regularly distributed on all images, most

  share in small values, and the average value of errors accumulated in large blocks is practically

  equal to 0. As a result, low-pass filtering is sufficient

  fisant to allow an efficient process, without transmitting-

  be no additional data. However, because

  accumulated errors are generated from the incomplete processing of compensation errors, errors

  accumulated are treated by the same process as the er-

  compensating rors through the four-shaft process

  ternary. Here, since the accumulated errors are hardly divided into smaller parts than in the case of compensation errors, and that these still have small values and an irregular distribution although they have been treated by the average value, low-pass filtering is performed after treatment with

the average value.

  Figure 4 is a detailed block diagram

  the motion estimator / compensator 52 and the general

  error rectifier / divider 53 shown in FIG. 3.

  Here, the motion estimator / compensator 52 includes a motion estimator 101 and first and second

  motion compensators 102 and 103. The general

  motion compensation error divider / divider 53

  comprises a buffer 107 and first, second and third

  sth subtractors 108, 109 and 110.

  When the motion estimator / compensator 52, shown in Figure 4, is in operation, the motion estimator 101 generates motion vectors, the first motion compensator 102 generates a first motion compensated image using the motion vectors emitted by the motion estimator 101 and referring to the previous original images, and the second motion compensator 103

  generates second images compensated in terms of motion

  using the motion vectors emitted by the es

  motion timer 101 with reference to the images

  constructed from previous original images.

  When the motion compensation error generator / divider 53 is in operation, the buffer 107 stores the second images compensated in terms of motion emitted by the second motion compensator 103, and constitutes reconstructed images, the

  first subtractor 108 generates compensation errors

  sation by calculating the difference between the pre-

  better motion compensated images

  by the first motion compensator 102 and the ima-

  current original ges, the second subtractor 109

  generates motion compensation errors in cal-

  being the difference between the compensated image in terms of motion emitted by the second motion compensator 103 and the current original images, and the

  third subtractor 110 generates the accumulated errors

  by calculating the difference between the er-

  motion compensation errors emitted by the second subtractor 109 and the compensation errors emitted

by the first subtractor 108.

  FIG. 5 is a flowchart intended to re-

  present the operation of the first and second pro-

  stopper with quaternary tree 54 and 55 represented on the

figure 3.

  In step 151, images of errors of a di-

  predetermined block dimensions are introduced, then, in step 152, an average value "M of a complete block and

  the mean values M1, M2, M3 and M4 of four sub-

  blocks divided by a quaternary tree

are calculated.

  In step 153, it is decided whether or not the

  mean value "M" is smaller than a predetermined threshold-

  mined "TH". In step 154, it is decided whether or not the average values M1, M2, M3 and M4 of the sub-blocks are smaller than the threshold "TH" when the average value

  "'is smaller than the" TH "threshold. When the values

  their MI, M2, M3 and M4 averages of the sub-blocks are all

  your smaller than the "TH" threshold in step 154, the

  paration by means of the four-tree process

  is stopped, and the quaternary tree structures memorized until now are transmitted to the

  multiplexer 56 of Figure 4 (step 156).

  At the same time, when the average value "M"

  is greater than the threshold "TH", in step 153, the value

  their mean "M" is subtracted from all the values existing in the block to generate a new value of the block (step 156) and generates new mean values M1, M2, M3, M4 of the sub-blocks (step 157). In step 158, it is decided whether the average values MI, M2, M3, M4 of the sub-blocks generated in step 157 are

  smaller than the "TH" threshold. When all the

  their averages MI, M2, M3, M4 of the sub-blocks are smaller than the threshold "TH", the division carried out by the process by quaternary tree is interrupted, and the

  quaternary tree structures memorized until pre-

  sent are transmitted to the multiplexer 56 of the figure

4 (step 155).

  However, when at least one of the average values M1, M2, M3, M4 of the sub-blocks is greater than the threshold "TH" in step 158, it is decided whether a block

  current is greater than the minimum block (step 159).

  When the current block is larger than the mini block

  mum in step 159, the block is divided by tree quater-

  nary, the quaternary tree structure is memorized, and then we return to step 152, and another treatment by quaternary tree is repeated for each

sub-block (step 160).

  Figure 6 is a detailed block diagram

  an adder / filter 64 shown in FIG. 3.

  Here, first and second adders 201 and 202, a

  part 203 of coefficient determination / wire taking-

  tre, and a low-pass filter 204 are arranged.

  During the operation of the addition

  neur / filter 64 represented in FIG. 6, the first ad-

  editor 201 adds compensated images in terms of

  movement emitted by the motion estimator / compensator

  ment 52 of figure 4 to the compensation errors of

  production from the first quad-shaft processor

  inverse ternary 62 of Figure 4, and the second addi-

  actuator 202 adds the output from the first addi-

  operator 201 with accumulated minor reproduction errors

  its from the second quaternary tree processor in-

verse 63 of figure 4.

  The part 203 for determining the coefficient

  cient / filter intake determines the number of intakes

  function of the size of the sub-blocks of accumulated errors

  Reproduction lines emitted by the second reverse quaternary tree processor 63 of FIG. 4, determines a filtering coefficient by dispersion of images of the same size and located in the same position in reconstructed images emitted by the first adder 201. The filter low pass 204 filters the output from the second

  adder 202 using the filter coefficient deter-

  undermined by the part 203 for determining the coefficient

cient / filter plug.

  Figures 7A to 7I show examples

  schematically presenting the structure of a processing relating to accumulated errors and compensation errors. Figure 7A shows original signals, and Figure 7B shows accumulated errors. The

  accumulated errors (Figure 7B) exist irre-

  regular on signal areas. Figure 7C shows

  accumulated errors (see Figure 7B) combined with if-

  original original (see Figure 7A). When a filtering pass

  low is carried out on the signal of figure 7C, a reconstructed signal, as indicated on figure 7D, can be

  generates. As a result, the reconstructed signal very if-

  milaire to the original signal can be generated by unique-

  the low-pass filtering, because the average value of the accumulated errors is equal to 0 and because the

  accumulated errors are irregularly distributed.

  Figure 7E shows compensation errors

  station. The signal of figure 7F o the errors of com-

  existing thinking can be obtained when the compensation errors (see figure 7E) are combined with the original signal (see figure 7A). When the signal of figure 7F is divided by quaternary tree (but in the case of a first signal, by binary tree), only the zone

  as in FIG. 7G o there is a com-

  thought is divided into small parts. The area includes

  so many errors can be replaced by the average value, and as a result, compensation errors

  reconstructed using the quaternary tree and the va-

  their mean are shown in Figure 7H. When the signal of figure 7H is subtracted from the signal of

  FIG. 7F, the reconstructed images as represented

  tees in Figure 7I can be obtained. Such a pro-

  transferred by quaternary tree relating to compensation errors

  This can be very effectively implemented, since the compensation errors exist mainly in a concentrated manner over a predetermined area.

  Figure 8 shows the result of a thread-

  experimental tracing of accumulated errors remaining after processing of compensation errors. The moving image data used is a "foreman" image, which is sampled at a frequency of 10 Hz and has a

  format of 352 x 288. Experience gives the result fil-

  by five simple two-dimensional sockets and one

  fixed coefficient value instead of adaptive filtering

  tif depending on the picture, therefore, in fact, if

  adaptive filtering is used, a more efficient result

  cace can be expected. As shown in the result

  experience, image quality can be improved

  3 to 4 dB level without special coding processing of

accumulated errors.

  As described above, in a method for coding a moving image and in a device adapted for this purpose according to the present invention, motion compensation error signals are divided into

  compensation errors and accumulated errors, the

  rors concentrated in the compensation errors are treated via a quaternary tree processing to code a mean value and the others

  small errors in compensation errors

  tion are filtered, and the accumulated errors are mostly filtered according to the characteristics, and

  an average value of a part of these is coded.

  Consequently, the production of binaries is reduced compared to the TCD process or the tree process.

  usual quaternary, hence the usefulness of the present in-

  vention to code the moving image by having a very

low transmission speed.

  However, effective coding of

  compensation facilitates the significant reduction of a pheno-

  leads to discrepancy between the boundary lines of an object,

  mosquito effect and wire blocking effect

  trage, so it is effective in improving the quality

  image clarity, by comparison with the TCD process or by

usual quaternary tree.

  Furthermore, according to the present invention, the data

  born are processed only in the spatial area without transformation of the area, so that it is possible to eliminate the processing delays caused by the transformation of the area, as in the TCD process. By

  elsewhere, the accumulated errors and the errors of

  thought can be dealt with through a

  general quaternary tree, so that a particular material

  culier, due to special handling of errors, is not

not necessary.

Claims (9)

  1. Method for coding a moving image characterized in that it comprises the steps consisting in: (a) generating motion vectors by carrying out a motion estimation relating to current original images with reference to previous original images, and generate motion compensated images with reference to said motion vectors and to   reconstructed images of the original images previously- your,   (b) divide the errors of compensation for   event generated after performing said step (a) in accumulated errors and compensation errors,   (c) treat said er-   accumulated rors and compensation errors divided during step (b) as a function of the characteristic of each error, (d) quantifying with respect to an average in each quaternary tree said accumulated and compensation errors processed during said step (c), and determining a quantization coefficient as a function of a predetermined quantization step, (e) coding in variable length the data quantified during said step (d), (f) storing the data transmitted from l 'step (e),   (g) determine the quantization step to be   to terminate in said step (d) as a function of the volume of said data stored during said step (f), (h) to reverse convert said data quantified in said step (d) into data before quantification,   (i) carry out a tree treatment four-   inverse of a block having an average value in a   predetermined sub-block of the data transmitted from the   said step (h), to generate compensation errors   tion and accumulated reproductive errors, and   (j) generate the first reconstructed images   tes from said compensation errors and said accumulated reproduction errors generated from said step (i), and said motion compensated images generated from said step (a), and then generate final reconstructed images by   through low pass filtering.   2. Method for coding a moving image according to claim 1, characterized in that in said step (c), most of said accumulated errors are   processed by said low-pass filtering, and only one   tie whose average value is not equal to 0 is processed ted by quaternary tree.   3. Method for coding a moving image according to claim 1, characterized in that in said step (c), the concentrated errors of said errors of   compensation are coded through said processing   by quaternary tree, but the other errors of small size is not coded.   4. Device for coding a moving image characterized in that it comprises: a first buffer (51) intended to memorize previous original images and current original images, a motion estimator / compensator (52) intended to generate vectors movement by carrying out an estimation of the movement of said original images   current with reference to said original original images   teeth stored in said first buffer, and intended to generate images compensated in terms of movement with reference to said generated motion vectors and to   reconstructed images of said original images previously your,   a generator / divider of compensation errors   motion (53) for generating motion compensation errors through said motion compensated images generated by said motion estimator / compensator (52) and   of said current original images, and intended to divi-   ser said motion compensation errors by er-   compensation and accumulated error rors, of the first and second processors (54, 55) with quaternary tree intended to carry out a processing by   quaternary tree so as to code the errors of com-   thought and accumulated errors emitted by said giant   motion compensation error generator / divider (53),   a switching part (56) for   mutate and transmit said motion vectors generated by said motion estimator / compensator (52) and   the data transmitted by said first and second processes   seurs (54, 55) with a quaternary tree, a quantifier (57) intended to quantify the   average value in a quaternary tree of fragmented data   its coming from the switching part (56), and desti-   born to determine a quantification coefficient based on   tion of a predetermined quantization step, a variable length coder (58) intended to perform variable length coding of said data quantized by said quantizer (57), a second buffer (59) intended to store the   data sent from said length encoder will   reliable, a bit rate control device (60)   intended to determine said quantification step at four-   define said quantifier as a function of the volume of data stored in said second buffer (59),   an inverse quantizer (61) for trans-   form said data quantified by said quantification   (57) in data before quantification,   first and second quad shaft processors   reverse ternary (62, 63) intended for milking   inverse by quaternary tree of the blocks with an average in a predetermined sub-block of said data   emitted from said inverse quantifier, for   generate compensation errors and accumulated errors   reproduction models, and an adder / filter (64) for generating first reconstructed images by adding said compensation errors and said accumulated errors of   reproduction issued by said first and second prooes-   sisters with reverse quaternary tree (62, 63) to said ima-   motion compensation issued by said estimate   motion compensator (52), and intended to generate   drer final reconstructed images by processing er-   rors irregularly distributed by filtering.   5. Device for coding a moving image   according to claim 4, characterized in that said es-   motion timer / compensator includes:   a motion estimator (101) for   generate these motion vectors by performing an   motion timation relative to the original images   rantes with reference to said original images previously- your,   a first motion compensator (102) for   intended to generate first motion compensation images using said motion vectors and with reference to said previous original images, and   a second motion compensator (103) for   aimed at generating second motion compensation images using said motion vectors and   reference to the reconstructed images of said original images previous nales.   6. Device for coding a moving image according to claim 5, characterized in that said ge-   generator / divider (53) of motion compensation errors   This comprises: a buffer (107) intended to memorize said second images compensated in terms of movement emitted.   by said second motion compensator (103) for generate reconstructed images,   a first subtractor (108) intended to generate   make compensation errors by obtaining the difference   rence existing between said first compensated images   in terms of movement emitted by said first compensation   motion detector (102) and said current original images,   a second subtractor (109) intended to generate   From said motion compensation errors in ob-   taking the difference existing between said second motion compensation images emitted from said   second motion compensator (103) and the images common ginal, and   a third subtractor (110) intended for   generate these accumulated errors by obtaining the different   rence existing between said motion compensation errors emitted by said second subtractor (109) and said compensation errors emitted by said first subtractor (108).   7. Device for coding a moving image   according to claim 4, characterized in that said ad- adder / filter (64) comprises:   a first adder (201) intended for addi-   manipulating said motion compensated images emitted by said motion estimator / compensator (52) to said reproduction compensation errors emitted by said first reverse quaternary tree processor (62),   a second adder (202) intended for addi-   the output of said first adder to said accumulated reproduction errors emitted by said second reverse quaternary tree processor (63),   a part (203) for determining the coefficient   cient / filter socket intended to determine intake of   filter according to a block size of said er-   accumulated reproduction reurs emitted by said second   reverse quaternary tree processor, designed to   complete a filter coefficient based on a dis-   persion of said reconstructed images emitted by said second reverse quaternary tree processor (63), and a low-pass filter (204) intended to filter the output of said second adder (202) using said filter coefficient determined by said part ( 203)   coefficient determination / filter intake.   8. Device for coding a moving image according to claim 7, characterized in that the number   of filter sockets of said part (203) for determining   coefficient / filter intake is proportional to the size of the sub-blocks of said accumulated errors   reproduction emitted by said second processor at ar- reverse quaternary bre (63).   9. Device for coding a moving image   according to claim 8, characterized in that in the-   said filter coefficient of said portion (203) of   coefficient termination / filter intake, the value of   coefficient in the central part of said pass filter   low is greater than other coefficient values   when the dispersion of said ter-compensated images   motion measurements, where compensation errors of the same dimension and located in the same positions as said sub-blocks are combined, is important, but the coefficient values are similar to each other when the dispersion of said compensated images in terms of movement is small.