FI94820C - Improved sub-sample pattern reset in an HDMAC decoder - Google Patents

Description

, 94820

The invention relates to a sub-sampling pattern restorer which is a combination of temporally subsampled fields in a decoder of a high-definition television receiver (HDTV) in a high-definition television receiver (HDTV) decoder.

In the European high definition television system, HDTV (High Definition 10 Television) is transmitted via satellite or cable using a MAC system. The HDTV studio standard has an aspect ratio of 16: 9 and comprises 1250 lines displayed at an interleaving ratio of 2: 1 and a field frequency of 50 Hz.

The picture is encoded when transmitted to a MAC compatible HDMAC signal by compressing the signal to its fourth part of the bandwidth in the Alenti v. Encoder so that the original picture can be restored in the receiver decoder. The line number of the transmission signal is thus half of the original, i.e. 625 lines, and the interleaving ratio is 2: 1. The input signal format for the bandwidth decoder is currently 1250/2: 1/50 Hz. In view of the receiver, the so-called A DATV signal that conveys information about the motion content of an image.

Compression is based on controlled subsampling controlled by motion information. This means that the less motion there is in the image, the higher the spatial resolution is used, and the greater the motion content of the image, the greater the temporal resolution at the expense of spatial.

In the above-mentioned bandwidth reduction encoder BRE (Bandwidth Reduction Encoder), the image to be transmitted is divided into blocks of 16 * 16 pixels, which are derived from one of the three image processing branches based on the motion content:?. Each signal applied to branch 30 has its own sampling pattern.

If there is no movement, the signal is controlled. An 80 ms branch where a subsampling pattern (decimation) is taken from each image field so that half of the pixels in the final image are transmitted. Samples are taken from each line and the decimated image forms a 35 quincunx sampling pattern. In this way, half of the compression is obtained and the other half is obtained by increasing the transmission time of one image from 40 ms to 80 ms. Haaran. · The name comes from here. Figure 1 illustrates the subsampling of an HDTV image in this branch and the placement of image samples in a 20 ms MAC field 2 94820. The upper four rasters describe the same HDTV image. Raster squares represent samples of a complete HDTV image. The image comprises an even and odd field, and for clarity, the complete image is presented four times in a row to make it easier to see where the samples in each MAC field come from. The samples of the first MAC-ken-5 come from the odd field of the high definition image (top left), where the black dots represent the original samples to be subsampled. As can be seen, every fifth sample (lines 1, 3, 5, etc.) of the odd original field is placed in every Fifth sample in MAC field 1. Follow but in the field, MAC field 2, placed from the even HDTV field on each line (2, 4, 6, etc.). ) every fifth 10 samples. Every Fifth sample of each line (lines 1, 3, 5, etc.) of the odd original field is placed in the MAC field 3 and these samples are between the samples to be placed in the MAC field 1. Similarly, samples entering the MAC field 4 are taken from the odd original field as every fifth sample taken between the samples to be placed in the field 2. The so-called encoder. with the "line shuffler" function, 15 samples to be placed in each MAC field are further transferred so that the samples of one line of one MAC field always come from two consecutive lines of the original HDTV field. As can be seen, MAC fields 1 and 3 together contain an odd HDTV field every other sample, and MAC fields 2 and 4 together contain an even HDTV field every other sample. Thus, it can be seen that four consecutive 80 ms mode MAC fields are required in the receiver's MAC decoder to recover the HDTV image.

If there are slow-moving objects in a part of the image, the video signal to be compressed is directed to a 40 ms branch, where the field transmission time is increased from 20 ms to 40 ;. To 25 ms and only every other quincunx subsampled field is transmitted, i.e. only every other sample is taken from every other field. Figure 2 illustrates in sampling the sampling and in which 20 ms MAC field the samples are placed. Samples of the odd field of the first HDTV image are placed in MAC fields 1 and 2 as shown, and samples of the odd field of the next HDTV image are placed in MAC fields 3 and 4, respectively. Thus, four consecutive: 40 ms mode MAC fields contain samples of two consecutive HDTV images. Thus, it can be seen that two consecutive MAC fields (fields 1 and 2 or fields 3 and 4) are required in the receiver decoder to recover the HDTV image. · 35

If there are fast-moving parts in the image, the video signal to be compressed is routed to a 20: 'ms branch, where the bandwidth is reduced by a quarter so that every fourth sample of each original image field is transmitted, i.e. there are samples from each line of the original 3,94820 image. This case is illustrated in Figure 3. Thus, MAC field 1 contains every fourth sample of the odd field of the HDTV image and MAC field 2 contains every fourth sample of the even field of the same HDTV image. Correspondingly, MAC field 3 contains every fourth sample of the odd field of the next HDTV image and MAC field 2 contains every fourth sample of the odd field of the same HDTV image. To restore the original HDTV picture, two consecutive MAC fields are required in the receiver encoder.

According to the information provided by the motion detector, a switch is controlled, which connects only 10 one branch at a time to the output of the BRE. This information, in which branch the output signal is generated, is encoded in a DATV signal, according to which at the receiving end the encoder directs the signal to the correct branch.

The receiver's MAC decoder has a compressed picture 625/2: 1/50 Hz returned to the original HDTV format, i.e. 1250/2: 1/50 Hz. The analog television signal received at the decoder is converted into video, data and audio signals. The decoder can be called a Bandwidth Restoration Decoder (BRD) and is analogous to the BRE described above. The BRD encoder will now be described in more detail with reference to Figure 1, and the description relates to the lu-20 minance signal. The received sample frequency is 13.5 MHz and the output frequency of the BRD is 54 MHz. The encoder comprises a "line deshuffler" block of 1.3-6 field memories 2 connected e.g. to a cascade, a subsampling pattern reset 3 and three interpolator branches 4, 5 and 6. Each interpolator branch produces a complete HDTV field and the output of the selected branch is switched by switch C BRD output signal.

:. 25 The position of the switch is controlled by the motion information contained in the received DATV signal. In block 1, two lines are generated from one incoming line so that each new line is formed from every other sample of the incoming line. After the deshuffling operation, the luminance signal is delayed in the field memories 2. The field signals are then processed in a sub-sample pattern return converter 3 SSPC (Sub-Sample Pattern 30 Converter), which is a combination block of temporally subsampled fields. It •: Reconstructs the original subsampling patterns. From the converter 3, the subsampling patterns are passed to branches 4, 5, and 6, where the original HDTV fields are formed by interpolating the missing samples. Controlled by the DATV information, switch C connects one of the branches at a time to the output of the BRD, thus obtaining the original HDTV image 35 in the format 1250/2: 1/50 Hz.

. * As stated above, the subsample pattern return SSPC combines consecutive set MAC fields. It compiles samples of the original image from different MAC fields and interpolates the missing samples when there is a temporal block edge in the queue of received MAC fields. Each branch has its own SSPC that produces the subsampling pattern it needs for the branch. Thus, the SSPC has two functions: normal operation and conversion operation. Normal operation means that if there is e.g.

S the 80 ms mode subsampling pattern to be processed, the SSPC of that branch has at its disposal the MAC fields corresponding to the mode from which the subsampling pattern can be retrieved. Because these fields are simultaneously available to 40 ms and 20 ms SSPCs, they process the subsampling patterns of these branches from these samples. However, these samples are not at the points where the 10 "correct" 40 ms and 20 ms samples should be, so the samples in the sample pattern to be formed must be interpolated from neighboring pixels. The sample patterns formed by the branches are thus not correct because their samples are "false", the samples belong to the transmitted 80 ms mode block. The 40 ms and 20 ms SSPCs are thus in conversion mode. The above means that always the SSPC of one branch is nor-1S in the paint operation while the SSPC of the other two branches is in the paint operation.

As stated above, it is the task of the SSPC to recover the subsampling patterns of the different modes. It combines the successive MAC fields it receives and assembles from them the subsampling pattern of the original HDTV image. In addition, it interpolates any 20 missing samples. The SSPC always generates a subsampling pattern field during the time it takes to receive one MAC field, i.e., four subsampling fields, also numbered 1-4, per 80 ms basic period (i.e., one HDTV picture), hereinafter referred to as HDTV field 1, HDTV field 2 , HDTV field 3 and HDTV field 4. It can be assumed that HDTV field 1 corresponds to 25 MAC fields 1, HDTV field 2 corresponds to MAC field 2, etc. For example, when the SSPC inputs have MAC fields 1-4 in Figure 1. , the SSPC processes the HDTV field 2 from these. If the next image sequence, the 2.40 ms mode block of the image, is received next, after the processing of the HDTV field 2 of the first sequence is completed, the SSPC inputs have the MAC fields 2, 3 and 4 and the MAC field 1 of Fig. 2, which already belongs to the next image sequence. The SSPC should further process the 80 ms mode block into HDTV field 3, even though not all 80 ms mode MAC fields of the same picture sequence block are available. Later, after 20 ms, fields 3 and 4 of Figure 1 and fields 1 and 2 of Figure 2 are input, but the SSPC should still process the 80 ms mode block into the HDTV field 35 4, even if not all MAC devices in the 80 ms mode block in the same picture sequence. fields are no longer available.

• 5,94820

Let us briefly consider the problems in reproducing the 80 ms branch subsampling pattern. The SSPC80 must produce a continuous data stream with an 80 ms interpolator branch in order for the interpolator to operate at all times and to avoid spatial edge errors. In the conversion operation of the SSPC80, the pixels that are missing from the sample pattern of the 80 ms branch are interpolated from the sample-5 patterns of the 20 ms and 40 ms branches. The interpolation of these pixels can be implemented with a 4-point median filter trim, because their interpolation result is only visible when the receiver is forced into one of the modes and when the interpolator mask exceeds the 80 ms block limit, i.e. when interpolating the Upper Edges of the blocks. Thus, samples from the sample pattern of 20 ms or 40 10 ms branches are fed to the four inputs of the filter, and the median of these samples is placed in the sample pattern of the 80 ms branch. Said mode constraint is a receiver testing method, so it is not appropriate to interpolate the conversion operation of the SSPC with a better interpolator than with a 4-point median filter. On the other hand, possible errors in the conversion interpolations with the 4-point median interpolator at the edges of the 1S blocks do not visibly affect the interpolation result of normal operation. In normal operation, when all the 80 ms fields required for processing a stationary HDTV image are available, no internal interpolators of the SSPC are required, but the interpolations are performed by an interpolator of the relevant signal processing branch.

20

Let us now briefly consider the problems in retrieving the subsample pattern of the 40 ms branch. The subsampling pattern is produced by two interpolators: SSPC40A and SSPC40C. The SSPC40A must provide a continuous data stream for the 40 ms interpolator branch so that the interpolator can operate at all times and avoid spatial edge errors when the interpo-25 generator mask exceeds the block boundary. The data stream produced by the SSPC40A is temporally identical to the data stream of the SSPC80, but the data stream produced by the SSPC40C is 40 ms behind the previous SSPC. Motion compensation needs samples from outside the block to be processed when the motion vector points outside the block, so the motion compensation block also needs a continuous data stream from both 30 to 40 ms interpolators. For motion compensation, the fields generated by SSPC40A and SPC40C can be interpolated at the temporal edge with a 4-point median interpolator in both normal and conversion operations. This can be done because the individual pixels generated in the motion-compensated field are difficult to detect due to the lower resolution.

35

Finally, we consider the problems in reproducing the 20 ms branch subsampling pattern.

. * In normal operation, the SSPC20 always has all the samples it needs available.

6,94820

A 4-point median interpolator can be used in the conversion operation because the resolution is so weak that no errors can be detected.

Crucial to the SSPC of the subsampling pattern restorer is how many S field memories are available. If four field memories are used, it means the absence of most samples at the temporal block edges. What is meant by temporal block edges is illustrated in Figure 5. It looks at the same block (sub-area of the image) with respect to time. The block is initially stationary, with the information placed in four consecutive MAC fields of the SO ms mode 10, represented by numbers within the block. The chronological order of the fields is from left to right. Each MAC field has its own unique field number, which is 1, 2, 3 or 4. In the next step, the block becomes motionful, so the information is placed in four consecutive 40 ms MAC fields 1, 2, 3 and 4. Then again, the block becomes stationary, so the information is in the 80 ms 15 mode MAC fields 1, 2, 3, and 4. Next, the motion content is large, so the block information is in four consecutive 20 ms mode MAC fields 1, 2, 3, and 4. The number below each field indicates which mode the field belongs to. Thus, successive field numbers 1-4 always contain samples of one block, as shown in Figures 1-3. Each block 1-4 of the four fields is associated with its own motion information, 20 received on a separate DATV channel, decoded and transmitted to the SSPC, which is able to return the correct subsampling pattern based on the DATV information.

When there are fields marked with an arc a at the input of the SSPC, the HDTV field 2 is processed to form an 80 ms 25 mode subsampling pattern of the SSPC.

3 ’This can be done because four consecutive fields of the same mode have to get wool. But when the input has fields marked with an arc b, only three consecutive 80 ms mode fields are available. When the input has fields marked with an arc line c, two 80 ms mode fields of the block to be processed are missing. In the cases shown by lines baa and kaa-K 0, the SSPC has to interpolate the missing samples. • It follows that image errors occur whenever switching from a motion-detecting mode to a stationary mode and vice versa, i.e. at the temporal block boundaries.

35 Table 6A in Figure 6 shows the problem cases when using four field memories in the decoder. In this case, samples of five in a row are available at all times; · From the manual MAC field: from the FM5, FM4, FM3 and FM2 memories, 7 94820 (4 pcs) readable fields and from the FM1 field currently being received. In the left column of the table, "FIELD 1 (or 2,3, 4) indicates the HDTV subsampling pattern field to be processed. (SSPC processes one HDTV subsampling pattern field for each received MAC field.)" RFN "refers to the MAC field at each input of SSPC 5. to the number (see Figures 1-3 and 5) and the notation "CASE" refers to the combination of input fields when processing each subsampling field, i.e., HDTV field 1, 2, 3, or 4. The bold RFN number indicates which HDTV sampling field corresponding to the MAC field is being processed. Cases where the temporal edge causes problems are marked in bold and in italics 10. The notations FM1 ... FM5 refer to field memories, the output of which gives each MAC field.It should be noted that FM1 is not really a field memory, but means at the moment The table is read as follows: select as starting position eg CASE 1. From memories FM5 and FM4 the readable MAC fields 3 and 4 belong in time to the above block and the readable fields of the other FM3, FM2 and FM 1 belong in time to the next block. Both blocks are stationary, so the MAC fields are 80 ms mode fields. The HDTV subsampling field to be processed is FIELD 1 and is associated with the field memory FM3 so that since the field memory has a MAC field 1, the HDTV field to be processed must also be 1, i.e. the number of the HDTV subsampling field to be processed. 20 represents the number of the MAC field in FM3. The temporal boundary of the blocks does not cause gelms because both blocks are stationary, so samples can be taken from both blocks. After one time has elapsed to receive one MAC field, i.e. 20 ms, FIELD 2 is processed (CASE 6). Compared to CASE 1, the outputs of the field memories have slid forward one memory, in column left-25, and the last MAC field 4 (RFN = 4) of the above-mentioned temporally block is now readable (i.e. just received) from the field memory FM1. There is no turnover difference in the temporal boundary between the blocks, so there is no problem in generating an HDTV field because the fields of both turnover can be used.

30. After another 20 ms, when the 80 ms mode HDTV subsampling pattern field 3 (FIELD 3) has to be processed, the situation is either CASE 9, CASE 10 or CASE 11, depending on whether the first MAC of the next block received from the field memory FM 1 is received. field (RFN = 1) 80 ms mode field (CASE 9), 35 40 ms mode field (CASE 10) or 20 ms mode field (CASE 11). In the latter two cases, there is a temporal turnover difference between the blocks, so not all fields are available in the 80 ms subsampling pattern return.

, 94820 After another 20 ms, when the 80 ms mode HDTV subsampling pattern field 4 (FIELD 4) has to be processed, the situation is either CASE 14, CASE 15 or CASE 16 depending on the state of the previous period, i.e. which mode field was received . CASE 15 and 16 are problem areas because there is a temporal-5 boundary between the blocks.

In this way, Table 6A can be used to view any situation at the SSPC inputs when using the four field memories. Cases CASE 1, 2, 3, 6, 9, 10, 11, 14, 15 and 16 are cases of normal operation from the point of view of SSPC80, 10 the rest, ie CASE 4, 5, 7, 8, 12, 13, 17 and 18 are konversiotapauksia. Since two fields in the same block of this mode are sufficient to return the 40 ms mode and 20 ms mode subsampling pattern, cases CASE 4, 7, 12, and 17 are normal cases for SSPC40 while the others are conversion cases. Cases CASE 5, 8, 13 and 18 are normal cases and other conversion-15 cases for SSPC20.

Thus, in the case of normal operation problems, which are indicated by the cases marked in bold in Table 6A, a stationary mode HDTV subsampling pattern must be formed, even though not all MAC fields are available.

20

In an HDTV receiver manufactured by the applicant using four field memories, the problem is solved as shown in Figure 7B. For clarity, Figure 7A still shows the problem cases in Table 6A, and Table 7B shows in each case the MAC fields used by the receiver algorithm in these cases to generate the HDTV sampling pattern. The algorithm is the median of four points. In cases CASE 2 and 3, there is a temporal block edge when moving from a motion-containing block to a stationary block. In the case of CASE 2, which moves from a 40 ms block to an 80 ms block, the last MAC field of the 40 ms block RFN4 is utilized in addition to the three 80 ms mode fields.

30 CASE 3 utilizes only three 80 ms fields. In the cases of CASE 15. and 16, which have a temporal block edge when moving from a stationary block to a motion-containing block, in the former case each field obtained from the memory is utilized, while in the latter case the first 20 35 ms mode block obtained from the field memory FM3 is utilized in addition to the three 80 ms fields. field. One disadvantage of an applicant-manufactured receiver utilizing four field memories is that it produces a luminance SSPC due to poor interpolator and inappropriate sample selection in a single block area at temporal edges with single isolated pixels and stamp errors from horizontal to 9,94820 lines and edges, as multiple temporal edges the missing samples are taken from the edges of the block from a neighboring block with different turnover.

Since this receiver uses two blocks with different motion content (80/40 ms blocks or 80/20 ms blocks) in two temporal edge cases, it causes more S error in the motion picture.

One natural solution would be to increase the amount of field memory to reduce problem cases. In most cases, all four 80 ms mode MAC fields in the same block would then be available. For example, the use of five field memories would improve the situation decisively, since there would only be such temporal edge problems in normal operation in two cases. The reference algorithm presented in the EU95 proposal for the HDTV development and implementation project in European countries is based on the use of five field memories. The algorithm is a weighted average. It can be shown that the use of five field memories leaves the cases CASE 15 and CASE 16 of Table 6A as problem cases, where the 80 ms mode changes to a 40 ms or 20 ms mode block. In this case, three 80 ms mode MAC fields of the same block are available. These cases are shown in Table 8A, where the notations used are the same as in Table 6A. The known Eureka-95 reference algorithm generates an HDTV subsampling field RFN4 in these two cases using samples from 20 of these three MAC fields. Table 8B shows the fields used by this reference algorithm. The sample set is therefore incomplete. The disadvantage of this algorithm utilizing five field memories is that it also produces individual isolated pixels and stamp errors at the horizontal edges on the horizontal lines and edges. In addition, it should be noted that the price of field memory is quite high.

The present invention provides an improved subsampling pattern restorer SSPC that does not have the disadvantages caused by the SSPCs described above at the temporal edges of the block.

The invention is characterized by what is stated in claim 1.

According to the invention, when the SSPC80 needs to generate a stationary mode HDTV subsampling pattern field in normal operation but some of the MAC fields of this block are missing, only the available mode MAC fields and 3 5 use the samples in these fields to generate the missing samples using an FMH (Fir Media Hybrid) interpolator. In this case, the samples to be interpolated are taken from the same turnover, i.e. from the same 80 ms mode block, in which case the interpolation result is 10 94820 good and does not depend in time on the magnitude of the movement of the previous or latter neighboring block mode.

According to a preferred embodiment, the interpolator mask comprises successive samples C, E and G from one line, successive samples A and B from the next line and successive samples D, F and H from the third line.

The function of the FMH interpolator is: X = MED [A, B, E, F, FIR, HOR, VER] where ΗΟΚ = (* ψ- 2 (A + B + E + F) (C + D + G + H) FIR = - - - 2 4 10 Samples A and B are from a different field than samples C, E, G, D, F and H. Samples A and B are successive known samples of the same line Samples C, E and G are successive samples of the same line and samples D, F, and H are consecutive samples of the next line.On the temporal edge, samples A and B are samples of an even field if the MAC field 4 is missing, but samples of an odd field if the MAC field 1. 15 is missing.

When all four fields are available in normal operation, or when the SSPC80 is in conversion operation, forming a subsampling pattern from samples of other mode blocks, it forms a subsampling pattern using a median of 4 points. The mask of the inter-20 polarizer comprises four immediate neighbor samples at the point to be interpolated, all of which may be from the same turnover or some may be from a different turnover.

The invention will now be described in more detail with reference to the accompanying schematic figures, in which Fig. 1 shows an 80 ms mode block subsampling, Fig. 2 shows a 40 ms mode block subsampling, Fig. 9 shows a 20 ms mode block subsampling, Fig. 4 illustrates a receiver MAC decoder, Fig. 5 shows incoming MAC fields, Fig. 6A shows cases of four field memory SSPCs where problem areas are marked, Fig. 6B shows cases of four field memory SSPCs when the invention has been used, Fig. 7A shows problem areas of normal operation of SSPCs in normal operation, Fig. 7B shows the fields used by the SSPC in one receiver, Fig. 8A shows the problem areas when using five field memories, Fig. 8B shows the fields used in another receiver, Fig. 9 shows the mask and function of a four-point median filter, Fig. 10 shows the FMH interpol used in the invention quota, Fig. 11 shows a subsampled HDTV image in one temporal edge case when moving to a motion block, Fig. 12 shows a subsampled HDTV image in another temporal edge case when moving to a motion block, and Fig. 13 is a block diagram of memories used in the SSPC.

20

Figures 1-8 are described in detail above and will be referred to below as appropriate. Reference is further made to Table 6A, which shows the problem areas of the SSPC when using the four field memories. The problem areas are, firstly, CASE 2 and CASE 4, when switching from 40 ms and 20 ms modes to 80 m mode, where only '25 three 80 ms mode fields are available and yet the first HDTV subsampling field of this mode should be formed. Problem areas include CASE 15 and CASE 16, where 40 ms or 20 ms MAC fields have already been received and four 80 ms mode fields are no longer available, but the last (fourth) HDTV subsampling field for this mode should be generated. .

30 In other words, the problem situations of normal operation are the temporal edges between blocks when the block mode changes from 80 ms mode to 40 ms or 20 ms mode or exits from 40 ms or 20 ms mode to 80 ms mode.

According to the invention, in the mentioned cases CASE 2, CASE 3, CASE 35 15 and CASE 16, only the three available 80 ms mode MAC fields are used to form the subsampling pattern and the missing samples are intupolated with an FMH interpolator (Fir Media Hybrid) and no samples are used for 40 ms. - or 20 ms mode fields, as in prior art solutions. In other normal operation cases- 12 94820 sa, when all four fields are available, i.e. cases CASE 1, CASE 6, CASE 9, CASE 10; CASE Ilja CASE 14, a subsampling pattern is formed using a 4-point median filter. In cases called the conversion operation, i.e. when a subsampling pattern is produced from samples of other modes (20 ms or 40 ms), S also uses said 4-point median filter.

Figure 10 shows the mask of an FMH filter used in the invention. The filter input samples are from the same HDTV image: consecutive samples C, E, G of the same line, two consecutive samples A and B of the next line, and consecutive samples D, F, and H of the third-10 line. The interpolated mLsel is denoted by X. The filter first calculates the average VER of the known samples above and below the pixel to be interpolated, the average HOR of the samples to the right and left of said pixel, and the value FIR with the FIR filter. The filter then determines the median of the above values and samples A, B, E and F 15. The filter function is thus: X = MED [A, B, E, F, FIR, HOR, VER \ where • HOR = (- ^ 2 (E + F) 2 {A + B + E + F) (C + D + G + H) FIR —- - - 2 4 «

To understand the operation of the filter, reference is now made to Figure 11. Consider the cases CASE 2 and CASE 3 in Table 6B, where MAC fields 1, 2 and 3 of the same block 20 are available, but field 4 is missing. This means that every second sample of the odd-numbered subsampling field is missing, i.e., as shown in Figure 1, the samples of the rightmost raster pattern are missing. If the samples in fields 1, 2 and 3 of Figure 1 are shown in the same image and the missing samples are marked with a question mark (?), A raster representing the under-sampled HDTV image in Figure 11 is obtained. Black ym-25 wheels are samples of an odd field (odd lines) and black squares samples of an even field (even lines). As observed, the missing MAC field 4 causes the absence of every second sample of the even line. The missing sample is described by a question mark. The filter window used in the invention is illustrated in Figure 11 by a rectangle delimited by a thick line. In it, the three consecutive samples C1, E1 and G1 in the top row correspond to C, E and G in Figure 10 and the associated formula. Samples B2 and F2 in the middle row of the window correspond to marks A and B in formula 5 and samples C3, E3 and G3 in the bottom row. correspond to samples D, F and H of the formula. The value X calculated according to the formula is placed in place of the missing sample (?) in the middle of the window. In this way, all the samples marked with a question mark in the HDTV image are calculated, after which a sub-sampling pattern corresponding to RFN field 4 is formed for transmission to the 80 ms interpolators.

10

Figure 12 also shows the sub-sampling raster of the HDTV image when the MAC field 1 is missing and fields 2, 3 and 4 are available. Again, the black circles represent samples of the odd line, the black squares the samples of the even lines, and the items marked with a question mark represent the missing samples of the even lines. The filter window of Mus-1S with this rectangle is shown and the interpolation of the missing sample (?) Is performed analogously to the description of Fig. 11.

When this FMH interpolator is used to recover the 80 ms mode subsampling pattern at the temporal edges, the individual noise peaks are effectively attenuated, which is a major advantage over the prior art weighted average interpolator. All interpolator input samples are from the same turnover, i.e. the same 80 ms block, so that this use of single block samples guarantees a more reliable interpolation result and thus does not depend in time on the motion content of the previous or next neighboring mode block. The FMH interpolator used • • keeps the lines and edges consistent and does not break them, as the prior art weighted average and 4-point median interpolator do. In normal operation, temporal edge errors are completely eliminated or greatly reduced depending on the image to be interpolated. This improvement is important because usually at the rest of the receiver, the image produced by the encoder is sharpened by silicon-30 peaking, resulting in temporal edge errors produced by the SSPC. become more visible.

Table 6B, which corresponds to Table 6A, shows, in different cases, the relative numbered RFN-labeled MAC fields used in each case to generate the HDTV subsampling pattern field. In cases 2, 3, 15 and 16, an FMH interpolator is used as described above.

14 94820

In other normal operation, when there are no temporal edges, but some of the fields used are from the next or previous block, which is also an 80 ms mode block, a median of 4 points is used as the interpolator. These cases are illustrated in Figure 6B by the cases CASE 1 and CASE 14. The filter mask is shown in Figure 5 9. The inputs are the immediate known neighboring samples of the interpolated pixel X in both the horizontal direction (samples A and B) and the vertical direction (samples E and F). In normal operation, when all four MAC fields in the same block are available, a subsampling pattern is generated directly from these samples and no interpolation is required. These cases are illustrated by the cases CASE 6, CASE 10 9, CASE 10 and CASE 11 in Table 6A.

The remaining cases in Table 6B show conversion operation cases where the SSPC80 generates an 80 ms mode subsampling pattern of 20 ms or 40 ms mode samples. In this case, the said median of 4 points is also used.

15

The invention is easy to implement in addition to the existing prior art subsampling pattern restorer SSPC, which uses four field memories. Figure 13 shows the memory arrangement of one such SSPC. The samples from the outputs of the four field memories FM1, FM2, FM3 and FM4, as well as the incoming samples just si-20, are fed to a combination of pixel delays and line delays, the outputs of which give 3 * 3 sample windows. Samples obtained from sample windows that are temporally spaced by the duration of one MAC field (20 ms) are controlled by control logic (not shown) to recover different subsampling patterns. The sub-blocks SSPC80, SSPC40A, SSPC40C and SSPC20 '25, which make different sampling patterns, are thus implemented with the same line and pixel delays, and the control logic directs the formation and derivation of the sampling patterns to the outputs (cf. Fig. 4). The implementation of the invention in this known arrangement requires only the addition of two line delays and two pixel delays after each. The delays to be added are shown in bold in Figure 11.

»M-i uin i.i i.at.

*

Claims (4)

A decoder for decoding a high-resolution video signal that is divided into image areas and coded in various ways according to the motion content of each area, the stationary image area at the encoding being encoded in four successive MAC fields, comprising: - a number of field memories (2) to which a MAC video signal in sample sequence form is transmitted, - signal processing branches each forming a complete high-resolution television image of the sub-sampling pattern applied, - a sub-sampling pattern driver coupled to the field memories, and said that is made up of samples from at least four successive MAC fields, and which forms samples of these fields by providing that the sub-sample pattern in the stationary image area is applied to the signal processing branch da inputs constituting all four successive MAC fields in which the stationary image area coded, 15. a first and a second interpolation means as ing in the repeater of the sub-sampling pattern, which interpolates the missing samples in the sub-sampling pattern to be fed to the signal processing branch, where among the four successive stationary image areas, one is replaced by a field from another image area, characterized in that the first interpolation hybrid is a (FMH) interpolator interpolating said missing sample from the samples into three stationary fields in the same image area where the replacement field is a field containing movement from another image area, - the second interpolation means is a median filter which interpolates said missing field sample where the replacement field is a field in another stationary image area and where fields in the stationary image area are not available. Decoder according to claim 1, characterized in that the filter function of it. first interpolator (FMH) is X = MED [A, B, E, F, FIR, HOR, VER] in which HOR 2 is 94820 ver = α ± ίΐ 2 FIR - (A + B + E + F) (C + D + G + H) 2 4 5 and in which A and B are immediately known neighboring samples to sample X to be interpolated on line n, C, E and G are three successive known samples on preceding line n-1, and D, F and H are three successive known samples on the following line n + 1. 10 Decoder according to claim 1, characterized in that the second interpolator is a 4 point median filter, whose filter function is X = MED [A, B, C, D] in which A and B are immediately known neighbor samples to the puncture to be interpolated. in horizontal direction and E and F are immediate known neighboring samples in vertical direction. 4. Decoder according to claim 2, characterized in that, in the direction of arrival of the video signal - a first, second, third and fourth field memory is successively provided, - to the output of the first, third and fourth field memory, line delays and pixel delays are arranged so that a 3 * 3 display windows may be formed, 25. To the input of the first field memory and the output of the second field memory, line and pixel delays have been arranged so that a 4 * 3 display window can be formed in which the samples come from four successive lines, whereby the decoder control logic controls the desired window samples for further processing.