FR2630283A1 - Device for temporal sub-sampling and motion-compensated temporal interpolation in an interlaced image sequence, use of such a device in the coding/decoding circuits of a high-definition television image transmission system; coding/decoding devices for such a system - Google Patents

Abstract

Device for temporal sub-sampling and temporal interpolation in an interlaced image sequence comprising a spatial prefiltering circuit, a spatial sub-sampling circuit, a temporal sub-sampling circuit, and a motion estimation stage. This device is useful in particular in an image transmission system comprising three processing branches 701, 702, 703 in parallel and termed non-compensatable, compensatable and fixed, the said motion estimation stage, a circuit 770 for making a decision based on the original image, on the images processed by the branches and on a displacement factor selected by the said stage, and a circuit 740 for routing the outputs from the branches according to the output signal from the decision making circuit. Application: high-definition television image receivers.

Description

DescriPtion ZDevice for temporal subsampling and time compensated temporal interpolation in a sequence of interlaced images, use of such a device in the coding and decoding circuits of a high-definition television image transmission system definition, and coding and decoding devices for such a system
The present invention relates to a device for temporal subsampling and temporal interpolation in a sequence of interlaced images divided into blocks of
I x J points (I, J positive integers) whose luminosity is expressed numerically. The invention also relates to the use of such a device in the coding and decoding parts of a high definition television image transmission system, as well as coding and decoding devices for such a system.

 The invention is in fact applicable essentially in the field of high definition television, in a system for transmitting high definition television images according to the MAC standard. It makes it possible to reduce the bandwidth of television signals with a view to their transmission, this transmission being ensured by means of an analog channel which conveys the data having undergone compression and to which is associated an auxiliary channel known as digital assistance. allowing the transmission of additional information relating to the movements of the images transmitted by the main analog channel.

 The problem which arises is in fact to maintain, in spite of the compression of information necessary to adapt the quantity of this information to the limited bandwidth of the transmission channel, the best possible spatial resolution, whatever the speed of movement of the content. images to transmit.

 An object of the invention is therefore to propose a device allowing a temporal subsampling and a temporal interpolation compensated in movement without appearing the blurring and the jerks traditionally observed due to the temporal interpolation in the presence of movement.

 To this end, the invention relates to a temporal subsampling and temporal interpolation device characterized in that it comprises a circuit for spatial pre-filtering of the sequence of interlaced images, intended to deliver a series of sequential images while limiting the bandwidth, a spatial subsampling circuit intended to reduce in each of said sequential images the number of samples, a temporal subsampling circuit, intended to halve the temporal rhythm of said sequence of sequential images , and a motion compensation stage on transmission, said stage itself comprising three image memories, provided in series for storing on the one hand the images of rank 2k and 2k + 2 to be transmitted and on the other hand share the images of rank 2k + 1, eliminated before transmission, and a device for estimating movement according to a method such as that of block correlation.

 Another object of the invention is to propose, in a high definition television image transmission system, a reception stage comprising a motion compensation device which appropriately uses the motion estimation carried out at program.

 For this purpose, such a reception stage comprises, in its decoding part, a signal processing branch composed of a motion compensation device itself comprising a spatial post-filtering circuit intended to perform spatial interpolation in the sequential images transmitted by the analog channel, then, in parallel on this circuit, a delay circuit and a motion compensation stage, and, to receive alternately the output of said delay circuit and that of said motion compensation stage, a switch intended to deliver a series of sequential images then supplied to a format conversion circuit for the transformation of this sequence into a series of interlaced images, then to a multiplexer delivering the sequence of interlaced images ready to be displayed, said motion compensation stage comprising it - even on the one hand two image memories in series for the storage of the images successively transmitted by the channel analog and postfiltered and on the other hand an adder to perform the half-sum of said transmitted images.

 Another object of the invention is to apply the principle and the implementation thus described in a high definition television image transmission system with several processing branches in parallel according to the greater or lesser importance of the movement in the original images.

The coding device according to the invention is, for this purpose characterized in that it itself comprises
(a) three processing branches, said to be non-compensable, compensable and fixed respectively, located in parallel, each receiving the original high definition image and each ensuring a spatiotemporal subsampling rate equal to 4
(b) after converting the original image into a non-interlaced format, said motion estimation stage
(c) a decision-making circuit on the one hand from the original image and from the images processed by said three branches and on the other hand from the displacement vector selected by said motion estimation stage, and also sent to a so-called digital assistance channel associated with said analog channel
(d) a circuit for routing the outputs of said branches according to the output signal of said decision-making circuit, said decision signal also being sent to the digital assistance channel associated with said analog channel in that said branches themselves same
(i) the first, or non-compensable branch, a first spatial filter followed by a spatial sub-sampler and a first format conversion circuit
(ii) the second, or compensable branch, a temporal sampler at rhythm 1/2, a second spatial filter, a spatial subsampler, and a second format conversion circuit
(iii) the third, a time filter, a 1/4 rhythm time sampler, a third spatial filter, a spatial sub-sampler, and a third format conversion circuit and in that said processed images received by said processing circuit decision making are those available respectively at the output of said spatial filters.

 Another object of the invention is also to propose, on reception, a decoding device using appropriately the processing operations performed on transmission in the coding device according to the invention.

Such a decoding device is then remarkable in that it comprises
(a) three processing branches, said to be non-compensable, compensable and fixed respectively, located in parallel and each receiving the transmitted image with a view to decompressing these transmitted images and rendering them in high definition format;
(b) a circuit for routing the outputs of said branches according to the output signal of the decision-making circuit, said signal having been transmitted via said digital assistance channel in that said branches themselves include
(i) the first, a dynamic interpolation circuit, a multiplexer, and a spatial post-filtering circuit
(ii) the second, an interpolation circuit
s dynamic, a spatial post-filtering circuit, an image reconstruction circuit from said displacement vector selected on transmission, and a switch for conversion into interlaced format, said displacement vector having been transmitted via said digital assistance channel
(iii) the third, a dynamic interpolation circuit, a multiplexer, a time filtering circuit, a spatial filtering circuit, and an interlaced format conversion circuit.

The features and advantages of the invention will now appear in more detail in the following description and in the accompanying drawings, given by way of non-limiting examples and in which
FIGS. 1a and 1b respectively show the coding parts, on the transmission side, and decoding, on the reception side, of an estimation and motion compensation device of known structure, in the case of an image transmission system of television
- Figures 2 and 4 show, respectively in the coding parts, transmission side, and decoding, reception side, an embodiment of a device according to the invention, in the case of a system for transmission of high definition television pictures
FIG. 3a shows in more detail an example of a motion estimation stage in the device of FIG. 2, FIGS. 3b and 3d explain the content of the estimation circuits of the stage of FIG. 3a, and the FIG. 3c shows an exemplary embodiment of the cells making up these estimation circuits
- Figures 5 and 6 show alternative embodiments incorporating the devices of Figures 2 and 4 respectively
- Figure 7 shows an exemplary embodiment, in a high definition television image transmission system, of a coding device according to the invention
- Figures 8a to 8c show the three processing branches of the device of Figure 7, and Figures 9a to 9c the corresponding image formats at the output of said branches
- Figure 10 shows the decision-making circuit of the device of Figure 7
- Figure it shows the decoding device associated with the coding device of Figure 7 in the described transmission system.

 The device shown in FIG. 1 in the case of application to a high definition television image transmission system comprises, in known manner, a coding part on transmission (FIG. 1a) and a decoding part on reception (figure lob), which cooperate to detect and estimate the movement inside the images to be transmitted and adapt the processing of image data to the greater or lesser importance of this movement. The images are captured here by a high-definition television camera (not shown) which analyzes the scene using an interlaced scan at 1250 lines and at the rate of 50 images per second. The camera naturally provides, after matrixing signals R, G, B, three types of signals: the luminance signal Y and the two color difference signals U and V (or chrominance signal). Thereafter, the description relates for example to the luminance signal but would apply just as well to the chrominance signal. It will therefore simply be specified that the output signal from the camera is sampled, and that the resulting samples are presented at the coding part of the delta input at a rate of 54 megahertz. As the transmission channel, in the case of the MAC standard, only tolerates a rate of 13.5 megahertz, a subsampling must be carried out before said transmission.

 It will be specified that the images could, at the limit, be processed point by point but that it is simpler to cut them into N blocks of m x n points. Instead of scanning points, the blocks are then scanned, to which N representative points of said blocks correspond.

 The motion estimate mentioned above is provided as follows. The coding part represented in FIG. 1a consists of several branches in parallel, three for example, called here 1, 2 and 3. These branches receive the samples formed as indicated above and each comprise a pre-filtering circuit 101 and a subsampling 102. Although the sampling structures are different from one branch to another, the subsampling rates are identical and here equal to 4. The filter templates are chosen so that the aliasing due to the sub-sampling is avoided. The outputs of the three sub-sampling circuits are sent to a routing circuit 103 which, according to the command received on a fourth input connection 105, ensures the selection of one or the other of said outputs for subsequent transmission via analog channel 10 of the transmission system.

 The command present on the connection 105 of the routing circuit, normally identical for all the points of a block, is determined by a decision circuit 106, according to a criterion generally linked either to a quantity measured from the image input, or the energy of the difference between the pre-filtered images, from circuits 101 for example, and the original image. In the first case (a priori decision), the quantity measured can be for example the movement or the speed of the objects present in the image and the decision is taken directly according to the value taken by this quantity. In the second case (a posteriori decision), the energy of the difference makes it possible to determine the branch which leads to the best image reconstruction using the transmitted samples, and therefore to carry out the referral accordingly.

The information delivered by the decision circuit 106 is also sent to a so-called digital assistance channel 20.

 Similarly, on reception, the corresponding decoding part, represented in FIG. 1b, firstly comprises a demultiplexing circuit 152 which, from the signal transmitted via the analog channel 10, supplies three post-filtration circuits in parallel 153 images with appropriate regular structures. Finally, a multiplexing circuit 154 receives the outputs of these post-filtering circuits and allows, from the multiplexed signal, the generation of an image that can be viewed on a screen 155 at high resolution. The signal transmitted via digital assistance channel 20 is supplied in parallel to circuits 152 and 154.

 The sub-sampling structures, which differ from one branch to another as we have seen, can be purely spatial or else allow the elimination of a certain number of images in the time direction. We then have a proportionally higher number of samples to represent the spatial contents of the images but, in return, in the event of movement, the temporal contents reflecting this movement are degraded.

 Two types of degradations can be more particularly noted on the reconstructed image: on the one hand the uniform movements are altered, the objects moving in spurts, on the other hand the resolution decreases suddenly as soon as a fixed object becomes mobile , due to pre-filtering and post-filtering. Both of these faults are visually annoying, and the presence of an estimation and of a motion compensation will remedy them very notably, by eliminating the jerks and by preserving the resolution for a range of speeds. wider.

 The principle of motion compensation according to the invention is as follows. In the sequence of images considered, one image out of two is firstly eliminated (that is to say the spatial information available at a given time t). Thus, if the cadence or temporal frequency is for example 1 / T where T corresponds to the time interval separating two successive images, the time interval after temporal subsampling will be 2T. By calling 2k-1, 2k, .2k + 1, etc ... the successive ranks of the original sequence of images, this means that the images associate for example with times t + (2k-1) T, t + ( 2k + l), etc ..., or images here of odd rank, are eliminated.

 In parallel with this elimination of images, movement information is determined by a motion estimation method providing for the allocation to each block of the images to be eliminated, here the odd images, a displacement vector D such as the error of reconstruction of the block is minimal. This movement information is then used on reception to reconstruct the eliminated images, each block being reconstructed from the average of the information of two consecutive images in the direction of movement associated with the block. We thus limit the defects due to the imprecision of the estimator, when this one used until now the traditional solution of a reconstitution of an odd frame (or respectively even) from the frame of opposite previous parity , a solution which had the drawback of distorting the contours of solid objects.

 The principle thus defined is applicable for example in a high definition television image transmission system and first of all in the coding part of the transmission stage of this system.

 In what follows, the spatiotemporal sub-sampling carried out makes it possible to obtain a sampling rate of 4 (2 in spatial, 2 in temporal). The motion estimator that will be used is based on the so-called block correlation method (in English: block matching method), with a search range equal to: horizontal displacement +3, vertical displacement +3). However, this choice is not imperative and should not limit the present invention, for the implementation of which an estimator of another type could absolutely be used. It will also be specified here that the branches 1, 2, 3 correspond to different speeds of movement on the images, the basic interval between images being, in the example described, 20, 40 and 80 milliseconds for the branches 1, 2 and 3 respectively. It is obviously in branches 2 and 3 that there is interest, the movement being slower than in the case of branch 1, to keep the best resolution. The invention is therefore applicable to branch 2, the extension to branch 3 being described below.

 According to the example shown in FIG. 2, the device according to the invention firstly comprises, on transmission, for the branch referenced 2, a space prefilter circuit 20.1 receiving the input image, which is an interlaced image 50 hertz, 2: 1, 1250 lines, 54 M-samples (= 54.106) per second. This circuit 201 makes it possible to obtain a sequential image, while limiting its spatial bandwidth in order to avoid aliasing due to temporal subsampling. Circuit 201 is followed, in series, by a temporal sampling circuit 202 which halves the temporal rhythm of the image (we therefore go to 25 images per second)., then a circuit 203 of spatial subsampling which makes it possible to reduce the number of samples in each image plane ( by example using a staggered line subsampling which removes every second point). At the output of circuit 201 is also provided, in parallel on the series connection of circuits 202 and 203, a motion estimation stage 204 which will now be described in detail.

 The objective of the motion estimation carried out by the stage 204 is to determine for each block of the image of rank 2k + 1 which is eliminated a displacement vector D such that one can obtain an approximation of said eliminated image at from the half-sum of the non-eliminated images which surround it, in this case from the half-sum of the images 2k and 2k + 2. This approximation is here expressed by the relation (1), given in the appendix like the other mathematical expressions which appear in the following description. In this relation, X denotes the current block of the image 2kil, D denotes the motion vector when the motion estimation has been applied to the images 2k and 2k + 2, and I the approximation of the intensity of the point X of the current block of the image 2k + l.

 This objective can also be formulated by expressing that we want to associate with each block X of the image 2k + l a vector Dx such that expression (3) is minimal (this expression, in which DFD comes from the corresponding English terms "Display Frame Difference", is the approximation error attached to the current block and equivalent, for this block, as indicated by expression 2), to the sum of the squares of the DFD approximation errors on all the points of the block ). This known principle of examining the correlation between blocks (described in particular in the article by J.R. Jain and A.K.

Jain, "Displacement measurement and its application in interframe image quince ', published in the journal IEEE Transactions on Communications, vol.COM-29, n012, December 1981, pages 1799 to 1808) is implemented, in the floor of motion estimation 204 described here, in two distinct but extremely similar stages.

 The motion estimation stage ~ 204, represented by way of nonlimiting example in FIG. 3a, comprises, for the execution of these two steps, a set 340 of two practically identical estimation circuits 300 and 350, as well as three serial image memories 341, 342 and 343. The estimation circuit 300, for example, comprises, as shown in FIG. 3b which gives a more detailed representation, on the one hand nine identical cells 301 to 309 determining nine distortions (or estimation error as defined by expression (3) or expression (2)) relating to the following nine displacements Dx, Dy = (2,2), (2,0), ( 2, -2), (0.2), (0.0), (0, -2), (-2.2), (-2.0), (-2, -2). These nine displacements are stored in a memory 345. Each of the cells 301 to 309 itself comprises identical elements, 311 to 331 for the first, 312 to 332 for the second, etc ..., and 319 to 339 for the ninth . These elements considered for example for the first of the nine cells include first of all, as indicated in FIG. 3c showing this first cell, an adder 311, which is intended to carry out the half-sum of the images of rank 2k and 2k + 2 , then a subtractor 321 and, in series with these two elements, a circuit 331 for squared elevation and summation. The adder 311 receives the output of the image memories 341 and 343, and the half-sum of the images 2k and 2k + 2 thus obtained is supplied to the positive input for example of the subtractor 321 which receives on its input of opposite sign the output of the image memory 342 storing the image of rank 2k + 1. The output of the subtractor 321 is supplied to the circuit 331 for squaring and summing, the output of which constitutes the output of the cell 301.

 The respective nine outputs of the nine cells 301 to 309 are then supplied to a distortion comparison circuit 344 which compares the nine distortion values thus originating from the nine cells and determines which one is the lowest. That of the nine displacements which leads to this minimum block distortion is called Dminl and, after extraction of the memory 345, is sent to the second estimation circuit 350. This circuit 350, represented in FIG. 3d, repeats exactly the same operations than the estimation circuit 300, but on nine other displacement values which are as follows: Dx, Dy = Dminl + (1), Dminî + (1.0), Dmînl + (1, -1), Dminî + ( 0, 1), Dminl + (, O), Dminl + (0, -1), Dminl (-1, 1), Dminî + (-t, O), Dminî + (-1, -1), and which are this time stored in a memory 395 receiving Dminl.

 The estimation circuit 350 comprises nine cells 351 to 359, themselves comprising elements 361 to 381 for the first, 362 to 3ss2 for the second, etc., and which, as before, are an adder (361 to 369 ), a subtractor (371 to 379) and a square elevation and summation circuit (381 to 389) provided in series. The outputs of the nine cells 351 to 359 are supplied to a distortion comparison circuit 394 which determines the lowest distortion and allows the selection of the corresponding displacement Dmin2, that which minimizes the distortion for the current block X of the image 2k + l.

 The displacement thus selected is then sent to the digital assistance channel 20, while the output of the spatial subsampling circuit 203 is sent to the routing circuit 103. On reception, the motion compensation device receives a image sequence with a frame rate of 25 images per second, these images being spatially undersampled, and reproduces a sequence of images of 50 images per second, 1250 lines, 2: 1 interlaced format, 1440 dots per line, and 54.106 samples per second.

 This motion compensation device on reception comprises first of all, as indicated in the embodiment of FIG. 4, a spatial postfilter circuit 401 performing spatial interpolation to obtain a sequence of 25 frames per second, 1250 lines per image, 1440 points per line. This circuit 401 is followed by a delay circuit 402 (the delay provided here is 20 milliseconds), then by a switch 403 which allows, from two image sequences each having a period of 40 milliseconds but offset by 20 milliseconds, to reconstruct a sequence having a period of 20 milliseconds. At the output of the spatial post-filtering circuit 401 is also provided, in parallel on the series connection of the circuits 402 and 403, a temporal interpolation stage compensated in motion 404.

 This stage 404 comprises on the one hand two serial image memories 441 and 443, which store the two images successively transmitted by the analog channel 0 and postfiltered, that is to say the two images of rank 2k and 2k + 2 respectively, and on the other hand an adder 444 which is placed at the output of said memories and which makes it possible to perform the half-sum of the images transmitted, according to expression (4) where X represents the coordinates of the current point, Dmin2 the displacement attributed to. point and delivered by the digital assistance channel 20, I (X-Dmin2, 2k) and I (X + Dmin2, 2k + 2) the intensity at the points associated with X on the transmitted images of rank 2k and 2k + 2 respectively (taking into account the estimated movement), and i the intensity at point X of the eliminated image to be reconstructed. The output of the adder 444 constitutes the second input of the switch 403.

 The switch 403 therefore receives on the one hand images 1: 1 which are the images 2k, 2k + 2, etc ... transmitted every 40 milliseconds and on the other hand images 1: 1 which are the images Î estimated d after the displacement transmitted by channel 20 and whose period is also 40 milliseconds with a shift of 20 milliseconds relative to the transmitted images, and it therefore delivers a sequence of images having a rate of 20 milliseconds between images. A format conversion circuit 405 transforms this sequence of sequential images into a series of interlaced images, which is then supplied to a multiplexer 406. The output of this multiplexer 406 constitutes the high-definition interlaced image sequence ready to be viewed.

 Of course, the present invention is not limited to the embodiments described above and shown, from which variants or improvements can be provided without departing from the scope of the invention.

 The device structures previously described can indeed be modified in order to obtain a higher temporal subsampling rate, for example equal to 4. These modified structures are represented in FIG. 5, for the transmission part, and in FIG. 6, for the reception part.

 The device for estimating movement on transmission comprises, as indicated in FIG. 5, on the one hand the device of FIG. 2, here designated by the reference 510, and on the other hand, at the input of this device 510, a time-keeping filtering circuit, composed of a delay circuit 501 imposing a delay equal to T and of an adder 502, and a circuit 503 of time sub-sampling. The filtering circuit (501, 502) receives the 1250 line, 50 Hz, 2: 1 images and provides at its output a series of sequential images (1250 lines, 50 Hz, 1: 1) whose sampling makes it possible to obtain at the input of the motion estimation device 510 a series of sequential images 1250 2, 25 Hz, 1: 1. This sequence of images is processed by the device 510 which finally delivers a series of sequential images of 80 millisecond period.

 Similarly, on reception, the movement compensation device comprises, as indicated in FIG. 6, on the one hand the device of FIG. 4 designated here by the reference 520 and on the other hand, at the output of this device 520 , an interpolation circuit composed of a delay circuit 521 imposing a delay equal to T and of a switch 522. This interpolation circuit (521, 522) makes it possible to transform the series of sequential images 1250 lines1 25 Hz , 1440 points per line present at the output of the device 520 in a sequence of interlaced images having a temporal cadence of 20 milliseconds, that is to say in a series of images 1250 lines, 50 Hz, 2: 1, 54 M-samples per second, which is supplied to the multiplexer whose output constitutes the sequence of high definition interlaced images ready to be viewed.

 Furthermore, it has been seen above that the invention primarily relates to branch 2, known as the 40 millisecond branch, but it is also possible to apply the principle and the implementation described in a transmission system including such a branch and constituted as follows. FIG. 7 shows a possible embodiment, in a high definition television image transmission system, of a coding device in accordance with the present invention, FIG. 11 correspondingly showing the decoding device associated with this coding device in the transmission system.

 More specifically, the coding device of FIG. 7 firstly comprises, in parallel, three branches 701, 702, 703 called respectively non-compensable branch, compensable branch, and fixed branch. These three branches 701 to 703 each receive, on their common input E, the high definition image, noted 1250 lines, 50 Hz, 2: 1, 1440 dots / line, and consisting of a sequence of interlaced images, constituted as indicated now.

 In the so-called compensable branch 702, represented in FIG. 8b, the processing described above is carried out, executed in the case of FIG. 2 by the circuits 201 to 203 and, in the present case of FIG. 8b, by the circuits 721 , 722, 723. More precisely, a temporal sampler 721 at the rate 1/2 gives 1250 Q., 25 Hz, 2: 1, 1440 p./line images, which are then received by a spatial filter 722, then by a 723 spatial sub-sampler of rhythm 1/2 and staggered line delivering images 1250 Q., 25 Hz, 2: 1, 720 p./line. The output images of the spatial sub-sampler 723, conforming to the image format represented in FIG. 9b, are supplied to a format modification circuit 725 which, in two frames and in 40 milliseconds (therefore every 20 milliseconds) sends by groups of lines (1, 5, 9, 13, etc ... then 3, 7, 11, 15, etc ...) to an input of a routing circuit 740 whose function is specified below . The images present at the input of the spatial sub-sampler 723 (connection S2) are also sent to a decision-making circuit 770 which is described below.

 In the so-called fixed branch 703, represented in FIG. 8a, the sequence of interlaced images is firstly converted into a series of non-interlaced images in a time filter 731, of known type, which delivers images 1250 2. , 50Hz, 1: 1, 144Op./line, and which is followed by a time sampler 732 then by a spatial filter 733. This time sampler at 1/4 rhythm delivers 1250 images 2., 12.5 Hz, 1: 1, 1440 -p./line, which are then filtered spatially to avoid aliasing due to the subsampling performed. The output S3 of the filter 733 is sent on the one hand to another input of the decision-making circuit 770 , and on the other hand to a 734 spatial sub-sampler of rhythm 1/2 and staggered line, which delivers 1250 2, 12.5 Hz, 1: 1, 720p./line images conforming to the image format shown in Figure 9a, to be associated with Figure 8a. The output images of the spatial sampler 734 are then supplied to a circuit 735 known as format modification (in English: shuffle circuit) which, in four frames and in 80 milliseconds (therefore every 20 milliseconds), sends them by line groups (1, 5, 9, 13, etc ...,. then 2, 6, 10, 14, etc ..., and so on for 80 milliseconds) to the routing circuit 740.

 Finally, in the branch 701 said to be non-compensable, represented in FIG. 8c, the sampling structure is this time such that only one point in four is retained. The branch 701 comprises a spatial filter 711, then a spatial sub-sampler 712 of 1/4 rhythm and in staggered row, which provides 1250 images 2, 50 Hz, 2: 1, 370 p./line. These output images from the spatial sub-sampler 712, conforming to the image format represented in FIG. 9c, are supplied to a format modification circuit 715 which, in two frames and in 40 milliseconds (therefore every 20 milliseconds), sends to a third input of the routing circuit 740 first the samples located in a zig-zag on the broken line denoted 1 then those located also in a zig-zag but on a broken line denoted 2. As before, the images of input from the sub-sampler 712 (connection S1) are also sent to a third input of the decision-making circuit 770.

 From the foregoing description, made with reference to FIGS. 7, 6a to 8c, and 9a to 9c, it follows that the switching circuit 740 admits on its three inputs, noted respectively 741 to 743, three sequences of images which are compressed images, since, in each of the three branches 701 to 703, a number of image points have been eliminated. It will further be noted that, in each of the three sequences thus formed, the images contain the same number of points or samples to be transmitted in 20 millisecond increments.

 The routing circuit 740 then provides on its output a sequence of points or samples in which the content corresponding to each block of the original images will come from one or the other of the three branches according to the value of a decision signal received on an input 746 of this circuit 740 and coming from the decision-making circuit 770.

 This decision-making circuit 770, described in the following paragraph, is preceded by a motion estimation circuit 760 similar to the motion estimation stage 204 of FIG. 2. This circuit 760 is itself preceded by d 'A spatial filter 750 for switching from an interlaced image format to a non-interlaced format. The purpose of circuit 760, like stage 204, is to determine, for each block (or set of M × N samples) non-interlaced images of a certain rank (for example 2k + 1 images of odd rank) which are eliminated before transmission, a displacement vector D.

More precisely, this vector D must be such that an approximation of the eliminated image of rank 2k + 1 can be obtained from the half-sum of the non-eliminated images of rank 2k and 2k + 2 which surround it, the DFD approximation error attached to each block being minimal (it was noted above that this research aiming to make the DFD error minimal was already described in previous documents, and that the example given was not as a preferred embodiment).

 The decision-making circuit 770 can now be described in detail. This circuit 770, shown in FIG. 10, comprises three parallel channels - each comprising a comparison circuit, a square elevation circuit, and a block-by-block summing circuit, the outputs of these three channels being sent to three corresponding inputs of a circuit 1040 for comparing distortions and selecting the branch index corresponding to the lowest of them. The first channel, corresponding to the non-compensable branch, first comprises a subtractor lOt1, which receives on the one hand the input image 1250 Q., 50 Hz, 2: 1, 1440 line and on the other hand, the output S1 of the filter 711 of the branch 701. This subtractor 1011 is followed by an elevation circuit squared 1017 then by a summator 1018 on each block, the output of which expresses the distortion relating to the non-compensable and measured branch block by block.

 The third channel, corresponding to the fixed branch, likewise comprises a subtractor 1031, which receives on the one hand, via a delay circuit 1032 intended to compensate for the delay introduced by the space-time filtering of the branch fixed 703, the input image 1250 a, 50 Hz, 2: 1, 1440 p./line and on the other hand, via a circuit 1039 for converting from non-interlaced format to interlaced format, the output S3 of the filter 733. This subtractor 1031 is followed by a square elevation circuit 1037 and a block-by-block summator 1038 the output of which expresses the distortion relating to the fixed branch.

 The second channel, corresponding to the compensable branch, also includes a subtractor 1021 receiving on an input the input image 1250 a., 50 Hz, 2: 1, 1440 p./line, and, on its other input, the image obtained as follows from the output S2 of the filter 722. This output S2 is first sent to a spatial filter 1022 to pass from an interlaced image format to a non-interlaced format. The output of this filter 1022 is supplied on the one hand to an input terminal of a switch 1026 and on the other hand to two serial memories 1023 and 1024 which respectively store the images of rank 2k and 2k + 2 successively transmitted These two memories 1023 and 1024 also receive the displacement vector D, determined for each block by the motion estimation circuit 760 in order to obtain an approximation of the eliminated image of rank 2k + 1 from the half - sum of 2k and 2k + 2 images.

This half-sum is carried out by an adder 1025 provided at the output of the two memories 1023 and 1024. The output of the adder 1025 is itself connected to the other input terminal of the switch 1026, the output of which, coming alternately the output of the filter 1022 or that of the adder 1025 to reconstruct an image in interlaced format, is then supplied to said other input of the subtractor 1021. This subtractor 1021 is followed, as in the two previous cases, by a circuit elevation squared 1027 then a block summator. by block 1028 whose output expresses the distortion relating to the compensable branch. The distortions, thus available at the end of the three parallel channels which have just been described, are supplied, as indicated above, to circuit 1040 which compares and selects them the weakest of them to send the corresponding branch index to the input 746 of the routing circuit 740. This branch index constitutes said decision signal which commands in the routing circuit 740 the selection of either samples of exit from branch 701 non-compensable, either those of exit from branch 702 compensable, or those of exit of branch 703 fixed, with this restriction that, if we note in the sequence of decisions, the presence of a fixed branch decision "isolated, the latter is forced to be ultimately a non-compensable branch decision". The selection thus made therefore controls the transmission of one of the three branch outputs.

 Conversely, on reception, the images actually transmitted will be processed in the decoding device of FIG. 11 with a view to reconstructing the original high definition images. This decoding device firstly comprises, for this purpose, as in the case of FIG. 7, three parallel branches 1701, 1702, 1703 which each receive the images actually transmitted and the outputs of which are received respectively on the inputs 1741 , 1742, 1743 of a switching circuit 1740. These branches 1701 to 1703 are also called non-compensable, compensable and fixed respectively.

 In the so-called compensable branch 1702, the sequence of the transmitted images is supplied to a dynamic interpolation circuit, comprising a circuit 1721 for inserting zeros between the transmitted signals, as well as a circuit 1722, placed at the output of this last and introducing a delay of 20 milliseconds. This circuit 1721 generates from two successive sorts an image in the format of FIG. 9b, that is to say of cadence 40 milliseconds, in the non-interlaced format. This dynamic interpolation circuit is followed by an adder 1723 of the outputs of circuits 1721 and 1722 respectively. The image 1250 2., 25 Hz1 1: 1, t440p. Output line of adder 1723 is supplied to a circuit of spatial post-filtration 1724, then to an image reconstruction circuit comprising two memories in series 1725 and 1726 and an adder 1727 carrying out the half-sum of the outputs of these memories, according to the process already described above, with regard to the adder 444 of FIG. 4. The two memories 1725 and 1726 receive the displacement vector estimated on transmission and transmitted by the digital assistance channel 20. A switch 1728, which receives on the one hand the output of memory 1725 and on the other hand that of the adder 1727, finally delivers an image 1250 2., 50 Hz, 2: 1, 1440 p./line which is sent to the input 1742 of the switching circuit 1740.

 In the so-called fixed branch 1703, the sequence of images actually transmitted 625 Q., 50 Hz, 2: 1, 720 p./line is first supplied to a circuit 1731 called dynamic interpolation, intended to ensure as before a reinsertion of zeros between the signals actually transmitted for this so-called fixed branch. This circuit 1731 generates from four successive frames an image in the format of FIG. 9a, that is to say of frame rate 80 milliseconds, in non-interlaced format. In the image sequence thus obtained, the non-zero samples, in each image of the sequence, are located in a staggered line. These images 1250 Q., 12.5 Hz, 1: 1, 1440 p./ output line of the circuit 1731 are then supplied to a multiplexer 1732, then to a time filter 1735, then to a spatial filter 1736 at the output of which a 1250 Q, 50 Hz, 1: 1, 1440p./line image is available. Finally, a conversion circuit from non-interlaced format to interlaced format, comprising a delay circuit 1737, which delays said image by 20 milliseconds, and a switch 1738 controlled by a signal at 50 Hz, delivers a 1250 image., 50 Hz, 2 : 1, 1440 p./line, which is sent to entry 1743 of the 1740 switching circuit.

 The branch 1701 called non-compensable comprises, it, simply a dynamic interpolation circuit 1711, for the insertion of zeros as previously. This circuit 1711 generates from an input frame an output frame according to the format of FIG. 9c, that is to say a frame of rate 20 milliseconds or an image of frame rate 40 milliseconds in interlaced format. The branch 1701 then comprises a multiplexer 1712 and a circuit 1714 for spatial postfiltration which delivers an image 1250 Q., 50 Hz, 2: 1, 1440 p./ line then sent to the input 1741 of the routing circuit 1740.

 The output of the decision-making circuit 770, which, on transmission, had been sent to the routing circuit 740 is also sent to the digital assistance channel 20, with a view to restoring this information during decoding. . The routing circuit 1740, as well as the multiplexers 1712 and 1732, receive said output information from circuit 770 transmitted by channel 20. Circuit 1740 uses this decision signal to select correspondingly that of the outputs of branches 1703, 1702, or 1701 which is appropriate: the multiplexers 1712 and 1732 either simply deliver the output signal of the circuit 1711 or 1731 respectively if the decision signal corresponds precisely to the branch concerned (respectively the non-compensable branch 1701 or the fixed branch 1703), or deliver, in the contrary case, the output of the switch 1728 of the compensable branch 1702, by restoring as the case may be the format of FIG. 9a (case of the multiplexer 1732) or that of FIG. 9c (case of the multiplexer 1712). The reconstructed high definition 1250 Q., 50 Hz, 2: 1, 1440 p./line image is therefore finally available at the output of the 1740 routing circuit.

ANNEX

<SEP> 2
<tb> blocks
<tb>

Claims (10)

Claims 1. Device for temporal subsampling and temporal interpolation in a sequence of traced images divided into blocks of I x J points (I, J positive integers) whose luminosity is expressed numerically, characterized in that '' it includes a spatial pre-filtering circuit of the interlaced image sequence, intended to deliver a series of sequential images while limiting the bandwidth, a spatial sub-sampling circuit intended to reduce in each of said sequential images the number of samples, a temporal subsampling circuit, intended to divide by two the temporal rhythm of said sequence of sequential images, and a stage of estimation of movement on transmission, said stage itself comprising three memories d image, provided in series for storing respectively the images of rank 2k and 2k + 2 to be transmitted and on the other hand the images of rank 2k + 1, eliminated before transmission, and an estimation device motion ation using a method such as block correlation. 2. Device according to claim 1, characterized in that it is preceded, in series, by a time-keeping filtering circuit, comprising a delay circuit imposing a delay equal to the time interval T between two successive images of origin and an adder, and of a temporal sub-sampling circuit of the output of said adder, providing at the input of said device a series of sequential images of temporal cadence two times lower. 3. Transmission stage for a system for transmitting high-definition television images via an analog channel with limited bandwidth involving processing to reduce the amount of information to be transmitted, characterized in that it comprises, in its coding part, a temporal interpolation device according to one of claims 1 and 2. 4. A system for transmitting high-definition television images via an analog channel with limited bandwidth involving processing to reduce the amount of information to be transmitted, this system comprising at least one transmission stage and a reception stage and being characterized in that said transmission stage comprises in its coding part a signal processing branch composed of a device according to one of claims 1 and 2 and connected to an assistance channel digital transmission of information relating to said motion estimate to the reception stage. 5. Reception stage for a transmission system according to claim 4, characterized in that it comprises in its decoding part a signal processing branch composed of a motion compensation device itself comprising a spatial post-filtering circuit intended to carry out a spatial interpolation in the sequential images transmitted by the analog channel, then, in parallel on this circuit, a delay circuit and a motion compensation stage, and, to receive alternately the output of said delay circuit and that of said motion compensation stage, a switch intended to deliver a sequence of sequential images then supplied to a format conversion circuit for transforming this sequence into a series of interlaced images, then to a multiplexer delivering the sequence of interlaced images ready to be displayed, said motion compensation stage itself comprising two image memories in series p for the storage of the images successively transmitted by the analog channel and postfiltered and on the other hand an adder to effect the half-sum of said transmitted images. 6. Stage according to claim 5, characterized in that it is followed by an interpolation circuit comprising a delay circuit imposing a delay equal to T and a switch and intended to transform the series of sequential images of output of said device in a series of interlaced images at twice the rate. 7. Application of the device according to claim 1 to the embodiment, in a system for transmitting high definition television images via an analog channel with limited bandwidth involving processing to reduce the amount of information. to be transmitted, from a coding device characterized in that it itself comprises  (a) three processing branches, said to be non-compensable, compensable and fixed respectively, located in parallel, each receiving the original high definition image and each ensuring a spatiotemporal subsampling rate equal to 4  (b) after converting the original image into a non-interlaced format, said motion estimation stage  (c) a decision-making circuit on the one hand from the original image and from the images processed by said three branches and on the other hand from the displacement vector selected by said motion estimation stage, and also sent to a so-called digital assistance channel associated with said analog channel;  (d) a circuit for routing the outputs of said branches according to the output signal of said decision-making circuit, said decision signal also being sent to the digital assistance channel associated with said analog channel in that said branches include - same  (i) the first, or non-compensable branch, a first spatial filter followed by a spatial subsampler and a first format conversion circuit  (ii) the second, or compensable branch, a time sampler at rhythm 1/2, a second spatial filter, and a spatial subsampler, and a second format conversion circuit  (iii) the third, a time filter, a 1/4 rhythm time sampler, a third spatial filter, and a spatial sub-sampler, and a third format conversion circuit and in that said processed images received by said circuit decision-making are those available respectively at the output of said spatial filters. 8. A system for transmitting high definition television images via an analog band channel. limited bandwidth involving processing to reduce the amount of information to be transmitted, this system comprising at least one transmission stage and one reception stage and being characterized in that said transmission stage comprises in its coding part a device for coding according to claim 7. 9. In the reception stage of a transmission system according to claim 8, a decoding device characterized in that it comprises  (a) three processing branches, said to be non-compensable, compensable and fixed respectively, located in parallel and each receiving the transmitted image in order to decompress these transmitted images and restore them in high definition format  (b) a switching circuit-outputs of said branches according to the output signal of the decision-making circuit, said signal having been transmitted via said digital assistance channel in that said branches themselves include  (i) the first, a dynamic interpolation circuit, a multiplexer, and a spatial post-filtering circuit  (ii) the second, a dynamic interpolation circuit, a spatial post-filtering circuit, an image reconstruction circuit from said displacement vector selected on transmission, and a switch for conversion into interlaced format, said vector of travel having been transmitted via said digital assistance channel  (iii) the third, a dynamic interpolation circuit, a multiplexer, a temporal filtering circuit, a spatial filtering circuit, and an interlaced format conversion circuit. 10. Coding device according to claim 7, characterized in that the decision-making circuit comprises, in parallel, three distortion calculation channels between the original image and respectively the three processed images available at the output of said spatial filters. , and a circuit for comparing said three distortions and for selecting the branch index corresponding to the lowest of them.  Coding device according to claim 10, characterized in that, in said decision-making circuit, the selected branch index is changed to the index corresponding to the non-compensable branch when, in the decision sequence, there is in the presence of a decision isolated from the fixed branch index.