DE2304344A1 - METHOD FOR COMPRESSING THE MESSAGE FLOW IN DIGITAL IMAGE TRANSMISSION - Google Patents

Description

"Method for compressing the message flow in digital image transmission The invention relates to a method for message transmission with reduced Message flow preferably for the transmission of image information.

The transmission of analog television pictures requires a bandwidth of approx. 5 MHz. One is content with portraits when telephoning with a quality that a bandwidth of 1 Ntlz for the analog signals needed.

There is hardly any deterioration in digital communication the image quality when 7 to 8 bits are used per sample. this means taking into account the sampling theorem bit rates of # 70 Mbit / s for television quality and # 14 Mbit / s for portrait video call quality.

A telephone channel requires approx. 4 kHz or 84 kbit / s, so that one the following relation is obtained: a television picture channel analog = 1200 telephone channels 1 portrait picture telephone channel analog = 250 1 television picture channel digital> 1100 PCM ~ telephone channels 1 portrait picture telephone channel digital # 220 These numbers show the value of reducing the flow of messages when transferring images.

Is used instead of a transmission with pulse code modulation (PCM transmission) a transmission using differential pulse code modulation (DPCM), then approx. 3 bits per sample (L @ Stenger, A Method of Effective Intraframe -DPCM encoding, Message from the FTZ of the Deutsche Bundespost, Darmstadt (1971)). This simple one The method already delivers a compression factor of approx. 2.5.

A subsequent runlength coding (RLC) with adaptive working lower threshold in the DPCM device provides a resulting compression factor of approx. 5 (B. Wendland, F, May, An adaptive intraframe encoder for television signals, Conference report of the 20th annual BUG conference, Braunschweig, October 1972).

DPCM and RIC show the disadvantage of error propagation; hence became also proposed to switch between PCM and DPCM to avoid the influence of error propagation to reduce (Radio Mentor, Oct. 1972, p. 71).

The RIC is relatively easy to implement, but it just works almost optimal in terms of redundancy elimination.

You can only get a completely redundancy-free flow of messages through an optimal coding (BSTJ, J.27 (1948) pp 379-423 and pp 623-656 and MIT techn. Rep. 65 (1949)). This coding but is technically very complex and assumes an exact knowledge of the source statistics as well as a stationary signal source in advance.

Other redundancy-reducing methods have also become known, some of which come very close to the optimal code, (DAS 1 959 061) but small Exchange technical effort for loss of compression factor.

With each A / D conversion, not only redundancy but also information is suppressed, however, this is largely irrelevant information, so that here visually barely perceptible image errors arise. This is especially true for DPCDI and adaptive working methods. The art of channel coding is essential in the determination and elimination of irrelevance. DPCM from pixel to pixel reduces the message flow from 7 to 8 bits per sample to approx. 3 bits per sample. If one also forms the differences from image line to image line, then are possibly also 2 bits / sample value achievable (intraframe coding) (W.Thoma, Video Transmission Network with Intraframe DPCM and optional Interframe Coding, Procedings of 1972 International Conference and Communication Philadelphia, USA, June 19.-21., Pp. 39/1 - 39/6).

Since successive images are usually very similar, it can also be through the formation of the difference between the signals from image to image provides a considerable amount of redundancy and Irrelevance elimination can be achieved (BSTJ, Sept. 1969, 5. 2545-2553), however are these processes (frame-to-frame coding) as a result of the required image memory can only be realized at considerable cost, and this is where error propagation occurs particularly uncomfortably noticeable.

With the known compression method one has hitherto with acceptable Image quality can reduce the flow of messages to about a fifth.

The invention is based on the object in a method of Type mentioned at the beginning of increasing the compression of the message flow, the image errors by eliminating irrelevance and reducing the influence of the reproduction of Reduce transmission errors, with the resulting compression systems should work with as little hardware as possible. As digital computers are in increasing Dimensions should also be used as images in communication systems the image compression process can also be programmed as largely as possible on computers be. According to the invention, this object is achieved by the following process steps solved: The analog message signals are initially in a manner known per se by a device A according to the PCM-the samples are or DPCM-method binary coded and / in gray code, means B then becomes the "1" density minimized in the binary sequence, the binary sequence is divided into sequences and one Coding device C supplied, operating in parallel with different redundancy in the method encodes the sequences, only one of those in the coding device codewords of reduced redundancy that arise at the same time are sent out with With the help of a special character generator D, the Receiving side E communicated the way in which the binary sequence with reduced "1" density was generated and by which coding the transmitted code word was created, and after the digital signals have been transmitted over the transmission channel, the On the receiving side, the coding on the transmitting side is carried out through the associated reversal processes undone and in this way reconstructed the video signal.

This procedure results in a very high degree of compression of the message flow achieved with digital image transmission.

Advantageous improvements and refinements of the invention are described in the subclaims.

With the exception of a DPCM preliminary stage, compression is entirely digital carried out. The required processes are almost entirely on a digital computer programmable. The compression takes place in several successive stages, so that for lower demands the individual stages can also be used as simplified compression stages can work. In addition, the system is designed so that multiple use of the Building blocks can be easily achieved in multiplex operation, the invention is will now be explained in more detail with reference to FIGS. It shows: Figure 1 example of one video signal to be encoded.

Figure 2 reconstructed video signal of Figure 1, according to the invention Procedure.

FIG. 3 is a block diagram of the principle for coding at the transmission end the method according to the invention.

FIG. 4 is a block diagram of an exemplary embodiment for the transmission side Coding according to the method according to the invention.

FIG. 5 is a block diagram of an exemplary embodiment for the reception side Decoding according to the method according to the invention.

The first compression process is the well-known point-to-point DPCM technique used, but the co-coding with consideration for subsequent processes takes place in Gray code, which is known to have the property that neighboring numerical values differ by the smallest possible Hamming distance (smallest number of different Binary digits between the code words).

The coding is based on the principle of expansion of the threshold stock (communications technology Z. (1971), H.2, pp. 114-116): With DPCM, the difference compared to the previous signal value is digital coded. The Gray code words assigned to differences a become a Gray code words called. When using the Gray code, the code word set is determined according to the Table 1 divided into positive and negative difference values. For the difference values O to + 3, the assignment to the # -Gray codewords is unambiguous.

The quantization of the video signal can be linear or. For linear quantization, non-linear quantization / is shown in accordance with Fig. 1 the area between the black level and the white level of the video signal in FIG. 7 is positive Intervals and an O interval. The position of the uniqueness area in the code book according to Table 1 depends on the number of the interval in the the video signal ai-1 is or from the previous amplitude of the video signal. This position of the uniqueness area is determined by the following condition - ai 7 difference values Ai that are smaller than half the width of the quantization intervals are defined as zero and given the code word "000", other differences get the # -Gray codewords according to the uniqueness range, which by the video signal i-l 1 is defined. This description is based on that the video signal can only assume values between 0 and 7.

In the case of non-linear quantization, one proceeds as follows: The Quantization intervals are given the numbers under #i in the code book according to Table 1 stand. A distinction can be made between zero and 6 non-zero values for #i will . If SW is the peak value of the white amplitude and sb is the level of the porch is, the condition sb # ai-1 + #i # sw determines the position of the uniqueness area in the code book, exactly as with linear quantization. Dependent on on the non-linearity of the quantization and the previous level value ai 1 one obtains more positive or more negative quantization intervals depending on which one in the same way as explained for linear quantization. For one optimal quantization must depend on the width of the quantization intervals can be controlled by the previous level value ai-1 1. For each setting are to use a total of 7 quantization intervals each.

Uniqueness area (shifted up or down depending on a) # -Gray code 100 7 101 6 111 5 1 1 0 4 0 1 0 3 0 1 1 2 0 0 1 1 0 0 0 0 0 0 0 -0 1 0 0 -1 1 0 1 -2 91 1 1 -3 1 1 0 -4 0 1 0 -5 011 -6 O 0 1 -7 Table 1 FCodebuchJ PCM interpolation points for interrupting the error propagation must be marked as such so that the receiver can process them accordingly. In addition, the encoder must be switchable from DPCM to PCM or two encoders must be available. However, it is also possible to form supporting digits using the DPCM code # by forming the difference between the source signal and a fixed reference value - for example, the level "on - at a suitable point. The DPCM value then corresponds to the digitized absolute value of the signal at this point.

With DPCM a very coarse quantization is used, so that the supporting values can have a significant quantization error. But now, as a result of the Mach's phenomenon (J. Opt. Soc. Am., Vol.51, (July 1961) pp.740-746) such Rough quantization is permitted for signal jumps, i.e. not perceptible in the picture. The support points must therefore be introduced at signal jump points.

The signal jump points can advantageously be carried out simultaneously serve to identify the support points for the recipient: Stepping after a sequence of more than j zero words of the # -Gray code, i.e. after more than j equal level values m -code word / on different from zero, so if there is a jump in the signal, then the following analog difference value is compared to the named 'interpolation point value' fixed reference value, for example the zero level M, and as a DPCM Gray code word sent. The reference points are always on contours in the reconstructed image, so that the quantization errors are not noticeable.

You can of course also specify that a support point is always then is inserted when the signal jump exceeds a certain size, i.e. when after more than j zero words, m code words that differ greatly from zero occurred are. The goal of further compression is now to get the To thin out the binary character flow after the DPCM-Gray code conversion in such a way that few ones remain, so to minimize the "oil density".

The information content of a sequence is reduced by enlarging it the unequal distribution of ones and zeros, so that a subsequent optimal coding then results in the smallest code word set or smallest possible code word lengths, if the possible minimum of the 't density is present in the binary character stream.

The next step assumes that when there are many black-to-white transitions in images the signal differences of successive pixels are similar in size or are the same size and especially strong in non-technical images before and after Signal jumps, to which hard contours correspond in the image, approximately constant levels lie. This means that successive A-Gray code words mostly only differ by 1 bit or not at all. Therefore, the modulo 2 differences of successive A-Gray code words are formed.

The resulting code words are hereinafter referred to as A Gray code words called. This additional formation of the difference therefore results in a further reduction the "l '<- density in the binary message flow. A Bundles with increased 't1' density.

The formation of the transmission sequence is used to minimize the length of the bundle explained in more detail using an example.

In Figure 1, as a coding example, a video signal along a Line shown. The abscissa is adjacent pixels, the ordinate is that video signal corresponding to the brightness values of the image. Set the samples represent the ai values.

In Table 2, the sample values taken from FIG. 1 are in the column 1 noted. The Gray code words corresponding to the sample values are in column 2 listed.

Column 3 shows the differences between successive samples reproduced.

With the help of Table 1, these difference values are obtained A-Gray code words in column 4.

Table 2 1 2 3 4 5 6 7 8 9 10 11 12 ai1 G1 # 1 # G1x ## B1x # G1x G1 ai2 # 2 # G2 ## G2x Z1-Z2 1st image line Z1 2nd image line Sending side Z2 Receiving side Sending side 0 000 0 000 000 000 000 0 0 000 000 000 0 000 0 000 000 000 000 0 0 000 000 000 0 000 0 000 000 000 000 0 0 000 000 000 1 001 1 001 001 001 001 1 1 001 001 # 000 1 001 0 000 001 001 001 1 0 000 001 000 1 001 0 000 000 000 001 1 0 000 000 000 2 011 1 001 001 001 011 2 1 001 001 000 3 010 1 001 000 001 010 3 1 001 000 # 000 4 110 1 001 000 001 etc. 4 1 001 000 # 000 5 111 1 001 000 001 5 1 001 000 000 6 101 1 001 000 001 # 6 1 001 000 000 7 100 1 001 000 001 7 1 001 000 000 7 100 0 000 001 000 7 0 000 001 # 000 7 100 0 000 000 000 7 ° 000 000 # 000 7 100 0 000 000 000 7 ° 000 000 # 000 7 100 0 000 000 000 7 ° 000 000 000 7 100 0 000 000 000 7 ° 000 000 000 1 001 -6 011 011 011 3 -4 110 110 101 1 001 0 000 001 001 1 -2 101 001 000 1 001 0 000 000 000 1 0 000 000 000 1 001 0 000 000 000 1 0 000 000 000 Column 5 provides the modulo 2 difference of successive A Gray code words with the insertion of the interpolation points according to the rules described below.

The reconstruction of the A-Gray code words including support points is shown in column 6. The original Retrieve Column 2 Gray codewords as detailed in Column 7.

Column 8 contains the sample values of the second image line and column 9 the difference values corresponding to column 3.

The associated A-Gray code words are given in column 10.

Column 11 again represents the formation of the modulo 2 difference resulting code words.

Column 11 subtracted from column 5 modulo 2 yields the column 12, which represents the picture line differences.

The algorithm for the formation of those listed in column 5 as Interpolation points code words / reads: As soon as after å = 2 zero words of the A Gray code m = 1 code word different from zero occurs, for example in table 2, line 4, the support point is sent. The following A-Gray code word also comes original for transmission and only the second word after the interpolation point is again as the difference between the A-Gray code words.

This procedure shortens the ones from ones in favorable cases existing bundles around a code word. In the example mentioned, this procedure is used the "1" density reduced to 7:63.

So-called gray wedges often appear in natural images. These result in slowly increasing voltage values in the video signal. This results in isolated cases non-zero samples occurring at the DPCM stage. Through the A8 Gray code word formation this often results in undesirable duplications of the AL Gray code words in the transmitted binary sequence.

One way out of this difficulty is to fundamentally after a zero sequence of A-Gray code words after the first of An A-Gray code word forwards zero different # -Gray code word, i.e. at this one Position dispensed with the AL Gray code word formation. Exceeds the sample value difference # a predefinable threshold α, for example the value A = 1, then becomes one the next code word # is forwarded as an interpolation point. This is not the case, then ## - Gray code words are generated again.

This procedure results in the following code word sequences: Zero sequence of ## Gray code words, a non-zero hA Gray code word, an A-Gray code word.

Now we have to distinguish between two cases: 1. If A <ob, then follow again ## - Gray code words 2. If A> dD, then follow: a Gray code word as an interpolation point, a # -Gray code word and further ## gray code words.

For m = 2, the rule for avoiding duplication can be applied by the ## - Gray code formation can be simplified: occurs after # j zero words of the # -Gray code an N-1 to 1 then the following code word is sent as an A-Gray code word.

In order to guarantee uniqueness for the support points, the following must be used Agreement to be made: It is assumed that the support point is the same Has the same sign as the previous sample; this is automatically the case, if you select level 0 as the reference value for the interpolation points.

Successive image lines are usually very similar.

In the example according to FIG. 1, it is assumed that the second line has a flatter course of the signal function on the trailing edge than the first line.

The modulo 2-line difference is used to further reduce the "1" density of the AL Gray code words. The result is shown in Table 2, column 12. In the present case Example, the 111 density is reduced to 2:63, but all are ones now compressed to a bundle and thus especially for bundle or runlength coding suitable.

In the case of leather line difference formation, it is now checked whether this is a binary sequence which is better suited for a subsequent, redundancy-reducing coding is than the originally encoded line. If no more favorable coding is found, then the original line is used. A special character (1 bit per Line) the recipient receives notification of whether a difference or an original one coded line is used. By sending lines without creating line differences error propagation in the vertical direction of the transmitted image is prevented.

To simplify the technical system, it is advantageous for å the same number of binary digits must be used for each line of the image. But this continues assume that the originally coded line does not exceed a certain "1" density or at least has a suitable "1" structure. If this is not the case, then must a recoding take place, which reduces the density or at least a restructuring which generates ones.

The next process is therefore - if necessary - one Recoding in the following way as an example: 1 1 1 is replaced by 1 1 0 or 0 1 0 1 0 1 will be replaced by 1 0 0 1 1 0 will be replaced by 0 1 0 0 1 1 will be replaced by 0 0 1 This recoding is preferably carried out in the current line the formation of the line difference, with support points being excluded from the recoding should. During recoding, AL Gray code words with a high "1" density are replaced by Gray code words with a smaller "1" density whose corresponding analog value corresponds to the original AA Gray code word value is as similar as possible or which has the smallest possible error in the reconstruction of the analog signal. This results in quantization errors, which are further Lessening the irrelevance can be viewed. The error occurs first Line appears for signal jumps that are corrected by the interpolation points. failure however, the contours are barely perceptible. The sender determines which error arises from the recoding in the reconstructed signal, hence the recoding limited to uncritical cases on the sender side, In table 3 in column 1 is column 11 of the table recoded according to the given rules 2, which as well as columns 5 and 8 of table 2 in table 3 for a better overview were added, reproduced for the second line of the image. From the code word 1 1 0 of column 11, line 18 becomes the code word 0 as a result of the above-mentioned recoding 1 0 set. The modulo 2 differences of this recoded column 11 with the column 5 of table 2 provides column 2 in table 3. This column 2 or the column 12 of Table 2 represents the binary sequence that the redundancy reducing encoders is fed.

At the receiving end, image line 2 is shown in AL Gray, as it is is indicated in column 3 of table 3, from columns 2 of table 3 and the previously received column 5 of table 2, which shows the coded picture line 1 reproduces, by component-wise modulo 2-addition of these two on the receiving side existing coded columns recovered with the corresponding error, the resulting from the recoding.

The reconstructed A-Gray table 3 is obtained by component-wise modulo 2 addition of the rows from column 3 in table 3 5 8 11 1 2 3 4 5 6 ## G1x ai2 ## G2x ### G2x (Z1-Z2) x ## G2 #G #x ai2x from table 2 # sending side # # receiving side # 000 0 000 000 000 000 000 0 0 000 0 000 000 000 000 000 0 0 000 0 000 000 000 000 000 0 0 001 1 001 001 000 001 001 1 1 001 1 001 001 000 001 001 1 1 000 1 000 000 000 000 000 0 1 001 2 001 001 000 001 001 1 2 000 3 000 000 000 000 001 1 3 000 4 000 000 000 000 001 1 4 000 5 000 000 000 000 001 1 5 000 6 000 000 000 000 001 1 6 000 7 000 000 000 000 001 1 7 001 7 001 000 001 000 0 7 000 7 000 000 000 000 000 0 7 000 7 000 000 000 000 000 0 7 000 7 000 000 000 000 000 0 7 000 7 000 000 000 000 000 0 7 011 3 110 010 001 010 010 -5 2 001 1 001 001 000 001 001 1 1 000 1 000 OCO 000 000 000 0 1 000 1 000 000 000 000 000 0 1 Code words corresponding to column 4 in table 3. With the aid of table 1, the difference values # in column 5, table 3 can be recovered from these, and the supporting values ai corresponding to column 6 in table 3 can be reconstructed therefrom.

It can be seen that with the exception of the fourth from last word, which is replaced by the Recoding is slightly corrupted, the original values ai corresponding to the Column 8 in Table 2 were retrieved.

The example was chosen purely by chance in such a way that it was created by the formation of the line difference and recoding only a "1" in the binary sequence according to table 3, column 2 for the second 3-picture line remains, so a "1" density of 1:63 arises and the error is negligible in the reconstructed signal function.

By "thinning out the binary sequence, the message flow is still not been reduced. This can only be done through the subsequent, redundancy-reducing coding happen. For embodiment of the system this coding is in the / two parallel processes provided alternately - depending on the structure the line -to be used. The recipient is informed by a second special character, for example 1 bit per code word, communicated which coding was used.

The thinning primarily results in single bundles in the binary sequence. The well-known runlength coding is therefore suitable. The zero runs are recorded as Addresses coded; subsequent single bundles are transmitted originally in a known manner (D? -OS 2 046 974). It only remains to be agreed which "1" structures as Bundles are to be assessed. A bundle is defined as a sequence of ones and zeros, it always begins with a "1" and ends as soon as a specified number e of consecutive ones Zeros occurred.

For C> 0, the beginning of the address must be identified in further training of the DT-OS 2 046 974 application, a sequence of flTullen is sent in front of the address will.

Sequences of binary characters smaller than t'1 "density can be used if Binary sequences of constant length are optimally coded, in the sense that the for Binary sequence associated code word is completely redundant. Requirement for this The coding is that a maximum of e ones occur in the binary sequence. This type of coding is explained in more detail below.

The given binary sequences are n binary digits long.

Since only a maximum of e ones are allowed in a sequence, exist different sequences. A binary code of length k has 2k code words if the code is redundant. One now sets: It follows: The expression for k gives the minimum length of the binary code in which the binary sequences are to be mapped.

Based on the methodology of the theory of error-correcting codes, the binary sequence B is described as a single-column matrix EBl: By multiplying [B) by a coding matrix FC) of the form code words of length k are obtained as a single-column matrix fDj dh

This already specifies the coding rule you are looking for.

Now the further task is to determine that the by Eq. 4 a one-to-one mapping from [B] to [D] supplies.

EDJ is the vector sum of columns #i in [C] over GF (2) scalar multiplied by the values bi from [B], ie

with coefficients from GF (2) However, this means that at least 2e = d-1 arbitrarily selected columns in Eq. 5 must be linearly independent. The coding matrix of a linear systematic block code (WW Peterson, Error Correction Codes, MIT-Press, New York, John Wiley 1961) must meet the same condition if the associated code is to show a Hamming distance d. Such codes can be treated as algebra of polynomials with coefficients from GF (p). The binary sequence [B] is interpreted as a polynomial f (X) = b1Xn-1 + b2Xn-2 ... + bnX0 (equation 6), where bi are the components of the matrix IB7.

In polynomial notation, a so-called generator polynomial g (X) represents of degree k the coding matrix [C]. Since generator polynomial g (X) can e.g. B. after Bose-Chaudhuri be determined so that the generated code has at least the Hamming distance d; however, this means that arbitrarily selected ones in the coding matrix FC] equivalent to g (X) Columns of number (d-1) are always linearly independent. One educates with such a Polynomial% # fx P h (X) + S; (Eq. 7) then all residual polynomials differ r (X), the number of permitted binary sequences when dividing Xkf (X) by g (X) arise from each other. This provides a one-to-one assignment of the quantities £ Bj = f (X) given to r (X). The process according to Eq. 7 is in the simplest way by a feedback Shift register realizable (Telfunken-Zeitung Jg. 40 (1967), H.1 / 2, p. 62 ff.). The coefficients of r (X) are sent out as a redundancy-free code word.

The unambiguous assignment of the set [B] # f (X) to r (X) is at the receiving end from rrX) also [B] * f (X) with the help of the means of error correction unambiguously determinable. To reconstruct the sequence B, one grabs the transmitted Code word r (X) as a syndrome that corresponds to the rules of the error correction method from the reception.

word is first obtained by the decoding process. There is a Series of procedures in which the so-called error word can be extracted from the Syndrome is described (R.G. Gallager, Information Theory and Reliable Communication, John Wiley, New York, London, Sydney, Toronto 1968; Technical review, Ifr. 29 (1971), pp. 25 ff.).

Nuj the code word r (X) is treated as would it be a syndrome. The associated error word is calculated according to the known ones Rules and this is then identical to the original sequence [B] # f (X).

In contrast to the error-correcting coding, here the residual polynomial r (X) is sent out as a code word. If g (X) belongs to a tightly packed, error-correcting code (BSTJ., J.29, (1950), 5. 147-160) (e.g. Hamming code), then the number of possible residual polynomials r (X) is equal to the number of all permissible binary sequences fBj. The code words r (X) to be transmitted are then redundancy-free and equation 1 is fulfilled. Bose Chaudhuri codes (BCH codes) are usually only approximately densely packed, in this case it is ie some residual polynomials of the set r (X) do not occur in the coding. If such an unused residual polynomial is caused by transmission errors, the receiver can use this for error detection. If a generator polynomial that describes a J3CH code is used for coding, then the transmitted code words also have error detection properties.

BCH codes are almost densely packed and therefore the associated g (X) almost optimal for source coding. The process for coding [B] # f (X) in r (X) is therefore im The following optimal coding is called OC. If Binary sequences of the same length with a maximum of e statistically distributed ones in code words If the same length are to be coded, then the present method is optimal. However, since the code words occur with different probabilities the overall message flow in the sense of information theory is not free of redundancy. To eliminate the redundancy caused by different probabilities of occurrence, Shannon-Fano coding of the code words r (X) is required.

It should be noted that for e = 1 the runlength coding of the one described here Optimal coding is equivalent.

For the runlength coding, dense bundles of ones are to be aimed for. The smallest possible number of ones per binary sequence must be used for optimal coding TBi can be achieved.

With regard to the optimal code, the binary sequence is not used an image line is coded serially, but first the first binary digits one after the other of all b line AEGray code words, according to column 2, table 3 or column 12 Table 2, then the second binary digits and finally the third binary digits. In this way a spreading of the one-beam occurs.

A relief for the decoding process is obtained by using the Recipient notifies the number of ones in the original sequence by a special character was present.

However, this slightly reduces the compression that can be achieved.

In an advantageous development of the method, however, you can then for switch different "1" densities to different optimally selected codes and increase the compression again.

It is also conceivable, depending on the type of 311-der to be transmitted, as an alternative to work serially or with spreading and to the recipient by another special character to communicate the type of processing.

and all other cyclic codes The Bose Chaudhri codes # are also suitable for error bundle correction and can advantageously use this property can also be used for source coding.

A code with the generator polynomial g (X) allows the correction of Error bundles of length b.

One now forms ###### = q (X) + #### (Eq. 9) with the sequence g (X) of Source. On the receiving side, r (X) is separated into a shift register with a Feedback entered. Then the feedback is switched on, which the Register to divider register according to Eq. 9 makes. The register is now calculating Step by step ###### = qi (X) + ##### (Eq. 10) i r 1st, 2nd, 3rd, ... step At the register is connected to one of the known error bundle display devices (Bulletin ASE 61 (1970) 16, pp. 720-724). At step j, this device indicates that the original error word is in the register as rj (X).

The location of the error bundle in f (C) is given by j. There is f (X) reconstructed on the receiving side.

It will also be useful to use the code words r (X) or the RLC code words to protect against transmission errors by means of an error-correcting code. Therefor must then add additional - but only relatively low - redundancy to the binary flow be added; this can also be done by the OC. A simple protection against transmission errors with the help of a disturbance detector (NTZ, issue 2, (1969) P. 113 ff.) Can also be provided in the system in an advantageous manner: Transmission errors are the result of interference voltages and fluctuations in the transmission rate, which always lead to falsification of the binary signals on the channel. The disturbance detector recognizes these falsifications. Alternatively, there are now three ways of using of the interference detector to eliminate errors within the scope of the method according to the invention.

1. The disturbed code word is replaced by the corresponding code word of the replaces previous line. This is advantageous when OC has been encoded. If a redundant code is also used, the error bundle correction accordingly (Technische Rundschau, r. 29, (1971), 5. 25 ff.).

2. With RIiC coding, it is sufficient to identify the disturbed characters or groups of characters, which are within an error bundle to be replaced. It is advantageous to Completely replace the address and associated information points in which the Fault detector has given a fault indication.

3. You can also rigorously cut the entire line through the previous line Replace if a fault is displayed in one line.

In principle, video signal structures can always occur that are used for compression are unsuitable. In the present system, this means that neither the OC or the RLC can process the corresponding binary sequence. A way out is to take the transcoding even further until the desired "1" density is reached. But you can too rigorously the message in question suppress. In this case, instead of the relevant coded binary sequence a special character is sent out. If the recipient recognizes this special character in a Line, then this line is not fed to the D / A converter, but the previous one Line written again on the display tube.

In the transmitter this means that the previous line Z.-1 is saved must be until the relevant line Zi is coded, since for the following line Z i + 1 the difference to line Zi 1 must be formed.

For the described method for compressing video signals an exemplary embodiment is explained below with reference to block diagrams. This The example serves only to illustrate the effort and is therefore not included in executed in every detail.

FIG. 3 initially shows an overview of the compression method according to the invention given. The analog message signals are initially generated in a manner known per se binary-coded by a device A according to the PCM or DPCM method and in Gray code shown. The "1" density in the binary sequence is then minimized by means B.

The binary sequence divided into sequences is used by a coding device C fed in parallel with different redundancy reducing Method encoding the sequences. A switch S is indicated in Figure 3, that only one of the code words produced in the coding device C at the same time is sent out. Before the transfer is made with the help of a special character generator D informed by additional special characters of the receiving side z in which way the binary sequence with reduced t11fl density was generated and by which coding the transmitted code word was created. After the transmission of the digital signals over the transmission channel is on the receiving side E by the associated reversal processes the coding on the sending side is reversed and this way the video signal is received reconstructed.

In Figure 4, the coding method of the transmitting side is more detailed treated. The point-to-point difference is derived from the analog signals from source 1 formed as differential stage 2 by a conventional feedback loop. An analog-to-digital converter 3 with Gray coding supplies the A-Gray code words that are stored in shift memory 4 can be entered for an A-Gray code word.

A zero test logic 6 is at the parallel outputs of the memory 4 connected, which can for example consist of an AND element. This zero test logic indicates whether a zero word or a non-zero word from the analog-to-digital converter 3 was formed. A counter 7 is connected to the zero test logic, which stores the zero words registered and emits a signal to a delay circuit 8, which in turn with a delay ### sets the differential stage 2 to "O" as soon as after at least j zero words m different from zero er code words / occur in memory 4. On this Way arise the support points. The stage 8 is in each case a signal St to one Control unit, not shown here for a better overview, if a support point was formed. The device 6 also causes the memory to be reset 4, when the word 001 # # = 1 occurs, in order to double the 001 by the AL-Gra to prevent code formation.

The ## - Gray code words are created using the modulo 2 addition stage 5 following the memory 4 through the bridging line between the input of the memory 4 and the second input of the addition stage 5.

Interpolation points are sent when the memory 4 and level 2 to? IOI? were asked.

The zero setting of the memory 4 takes place when the grid point is transmitted twice in a row, as there is also an original A-Gray code word after the interpolation point followed by an AA Gray code word again at the second sample after a support point should be formed.

The binary-coded message of a picture line at the output of the addition stage 5 is received from a memory 9 and during the line return to a of the two shift line memories 10 or 13 alternately via the switches 11 and 12 handed over. In this process, an up-down counter 15 pays the ones in line. The previous line is in the other memory 13 or 10, so that by adding the memory contents via a modulo 2 addition stage 20 the Differences Zi # Zi-1, hereinafter referred to as # Zi Zi-1, of two successive ones of the lines can be formed, With this difference formation now pays the associated Counter 15 goes back one step for each "1" entered in the memory. Exceeds if it has its status "0", then there is no reduction due to the formation of the line difference of the "1" density and by repeated modulo 2 addition in the Adding stage 20 creates the original line Zi again. The code words Z. der Line i or # Zi Zi-1 are activated by appropriately controlling switches 21 and 22 given into the line memory 24. During the handover, Z. will be in the issuing Memory 13 or 10 entered again at the same time using switches 14 or 19, possibly by forming # Z.Z. @ Z i-1 with the help of the other memory 10 and 13. The zero detectors 16 and 17, which are connected to the counters 15 and 18, respectively are, give the control unit the message whether a ## - Gray code word line or a difference line # ZiZi-1 is sent out.

From the memory 24 get through the switch 25, which is only every third Bit of the shift memory 24 forwards the binary characters to the optimal encoder 29 and at the same time to the runlength encoder 27. The one connected to the optimal encoder 29 I: 1flDetektor 30 determines whether there are not more than e ones in the sequence to be coded otherwise it reports to the control unit that the sequence cannot be optimally coded is. A binary digit detector 26 is connected to the run length encoder 27. He determines whether the RLC word is not too is long to fail the control unit to communicate that the sequence is not RLC encodable.

Detectors 26 and 30 send false messages to the control unit from, then the corresponding line in the transcoder 23, which is at the output of the memory 24 is coded to reduce irrelevance. The recoded line is in the associated Memory 13 or 10 and stored in memory 24 and, as before, the original line, Coded in units 27 and 29 to reduce redundancy. If this is again unsuccessful, then the character generator 28 gives a special character, for example AV, in the transmission word memory 34 or 35, from which the broadcast is made to the panel. The control unit controls The special characters for 1st RtC coding or OC coding are also in the transmit word memories 2. Original line or difference line 3. Recoding or no recoding That The system described works with compression within an image (intraframe). A further compression is obtained if the picture-to-picture differences are added (Nacrichtentechn, Z. (1971), H.2, pp. 114-116.

Of course, this requires an image memory. There is only one point A further reduction in the "l" density should be aimed at by means of image differences in the area of binary sequences that are very similar, i.e. before an optimal coding. The image difference formation must therefore before or immediately after the line memory 24 inserted. The digital signals of a complete picture have to be saved will.

The difference can, however, be calculated line by line with the binary signals of the following image. The resulting difference line signals are then the encoders 27 and 29 are supplied. The difference is formed by modulo 2 addition in the adder 401 of the corresponding memory content for the relevant line. here error propagation is to be prevented by transferring the original support points.

If image difference formation is carried out, then it is expedient to be precise as with the formation of the line difference, to encode the picture difference only if the "1" density is not yet sufficiently reduced without this measure. Otherwise the binary sequence is transmitted without image difference formation.

In this way one achieves a further interruption of the error propagation from picture to picture.

A reduction in the storage capacity is obtained if the image difference is carried out on the already redundancy-reducing coded signal. But for this you have to the optimal coding can be dispensed with, since these binary sequences may be similar. encoded into completely dissimilar code words. That is when using the RIC coding not the case, since similar binary sequences are also assigned similar code words are. It is useful to identify the addresses of the information centers also to be formed by a Gray code, since information points are then close together receive similar addresses in consecutive lines. The addresses should are then in the code word in determined places, i.e. shorter information will be set / 7-utomatically extended by zero sequences. The image difference memory is located then following the RLC encoder 27. However, this does not result in a reduction of the message flow in bit / sec, but only a further "1t" density reduction. In order to take advantage of this, it is necessary to repeat the image difference signals Encoder 44 or 46 to encode RLC or OC.

In Figure 4, the options mentioned for the establishment of a Image difference level shown in dashed lines.

If you work with a memory for the full image content, then the organization is particularly easy. Suffice it to use the binary characters from the previous images modulo 2 by the adders 401 or 42 to the binary characters of the currently to be processed Image to add. The picture-to-picture difference is then coded to reduce redundancy. If you store the already RIC-coded image signals in memory 41, then you have to the image memory 41 can be RLC and / or OC-coded again.

This is through the encoders 44 and 46 at the output of the adder 42 indicated.

FIG. 5 shows an exemplary embodiment for the decoding of the the inventive method compressed image signals on the receiving side specified. A special character detector 50 recognizes the mentioned special characters and reports this to the control unit, not shown here for the overview.

The incoming code words are corresponding to the special characters decoded either in the RLC decoder 51 or in the OC decoder 52.

The following line memories 56 and 58 alternately take the Information Z. or A Zizi-1. This is entered into the memories 55 and 59. As soon as the line Zi + 1 or AZi + 1Zi is in the associated line memory, the formation of the line difference may be reversed by serial modulo 2 addition in adder 62.

If necessary, a recoding by the encoder 23 is in picture 4 undo. A transcoder 57, which is the inverse of the encoder 23, is at the outputs the memory 55 and 59 are switched and its content can be controlled via the switches 60 and 61 are stored back in the memories 55 and 59.

The lines with LA Gray code words are obtained from the memories 55 or 59 entered into the memory 64 and via a modulo 2 adder 65 to the memory 66 supplied. The memory 66 with Räckführung via the addition stage 65 receives the A-Gray codewords. The memory 66 must be set to "O" or it must the respective word in memory only to the DA converter 68 out and not via the return line. This can be done by opening the switch 71 can be achieved when the support point detector 67, for example as an AND element checks for zero words in memory 66, announces a support point or if after 7 j zero words a 001 # # = 1 occurs.

The A-Gray code words are converted into a digital-to-analog converter 68 in the A values, which result in the video signal via an integrator 70, are converted. A Synchronization character generator 69 at the input of the integrator takes care of the insertion of the line synchronization signals.

Their transmission on the transmission path is superfluous, because that Binary system works synchronously, so that the beginning of the line is simply counted from incoming binary characters can be determined chin. The beginning of the line can also be used be signaled by a special character, so that fluctuations in the line frequency can be regulated.

The switches in Figures 3, 4 and 5 are from a central Processor or control unit switched so that the processes mentioned in the operating sequence can run appropriately. By the letter St and its associated arrows is indicated from which module control signals are received or emitted and in which process this happens.

Zs does not describe the state of the switches, since the Switch positions result in a simple manner from the process described. The functions of the control unit result from the processes described.

The implementation takes place in the technology usual for computers.

The memories are all intended for series operation so that the can use commercially available LSI shift registers. A reduction in required clock speed can be achieved by going for the line processes Provides memory arrangements with parallel transfer of entire lines.

The possible arrangements are for the image-to-image difference formation the image memory is only shown in the block diagram of the transmitter. On the receiving side these memories can also be inserted analogously for the reversal process.

The levels for reducing the 1 density and for reducing redundancy Coding are of course also suitable for compressing other analog signals. The application for video signals is only in the foreground here as an example.

Claims (28)

Claims 1. Method for message transmission with reduced message flow preferably for the transmission of image information, characterized by the following Method steps: The analog message signals are initially known per se Way by a device A according to the PCM samples or DPCM method binary-coded, the / represented in Gray code, means B then becomes the "1" density minimized in the binary sequence, the binary sequence is divided into sequences and one Coding device C supplied, operating in parallel with different Redundancy-reducing method, the sequences encoded only one of the in the coding device codewords of reduced redundancy that arise at the same time are sent out with With the help of a special character generator D, the Receiving side E communicated the way in which the binary sequence with reduced "l" density was generated and by which coding the transmitted code word was created, and after the digital signals have been transmitted over the transmission channel, the On the receiving side, the coding on the transmitting side is carried out through the associated reversal processes reversed and in this way reconstructed the video signal. 2. The method according to claim 1 in.dem in a manner known per se analog video signals DPCM are converted1 characterized in that for binary representation of the analog samples the Gray code is used around that after a sequence by at least j consecutive zero words of the A Gray code followed by m the next code word from zero different code words as the A-Gray code word dem Orignial sample value of the video signal and that the video signal difference formed by setting the differential level to a specified value and with represented by the Gray code. 3. The method according to claim 1 or 2, characterized. modulo 2 that to reduce the "1" density in the binary sequence the / difference of consecutive # -Gray code words are formed within each image line, so that ## - Gray code words develop, 4. The method according to claims 2 or 3, characterized in that that a # -Gray code word is forwarded after a support point. 5. The method according to any one of claims 3 or 4, characterized in that that following a zero sequence of Gray code words for one different from zero A-Gray code word the next code word is an A-Gray code word and thus to this one Place on the ## - Gray code formation is omitted if the # -Gray code word associated sample difference A C α. 6.Verfahren according to any one of claims 1 to 5, characterized in that the assignment of the # -Gray code words to the difference values # is selected as follows: A -7 -6 -5 -4 -3 -2 -1 ° and # l 2 3 4 5 6 7 AG 001 011 - 010 110 111 101 100 000 and that in order to eliminate ambiguities the assignment of A-Gray code words as a function of the previous sample of the video signal level ail takes place according to the following condition: 0 # ai-1 + # i # 7 if the video signal only assumes values between 0 and 7 or sb # ai-1 + Çw if the video signal only assumes values between sb and s. 7. The method according to any one of claims 2 to 6, characterized in that that for the support points on the receiving side always the sign of the potential is assumed by the previous sample and that on the sending side the possibly further # words compared to the potential value erroneously formed on the receiving side are formed. 8. The method according to any one of claims 1 to 7, characterized in that that the number of ones in a line is determined by an up-down counter that all modulo 2 difference of the binary characters with the previous line is formed and it is determined by the up-down counter whether by the Difference formation reduces the "1" density in the line compared to the original line and that otherwise the original line by occasional nodulo 2 addition the binary characters of the previous line are restored to the difference line and is saved. 9. The method according to any one of claims 1 to 8, characterized in that that the binary characters of the original line or the difference line are simultaneous with Coded using a runlength coding method and an optimal code method and that preferably RLC words are output and only then OC words output if the RLC coding fails while the OC coding fails permissible code word generated. 10. The method according to claim 9, characterized in that a recoding of the line takes place, if neither by the cptimal encoder nor by the runlength encoder the binary characters of a line can be encoded into permissible code words and that the recoding takes place in such a way that Gray code words with several binary ones can be replaced by Gray code words with fewer binary ones, with the Gray codeword with fewer binary ones is selected than the original Gray code word is closest in terms of value and that Umcodierun.x at the points occurs, cause. the only permissible errors in the reconstruction of the video signals / 11. The method according to claim 10, characterized in that the recoding in the Way is done that Gray code words used with a low "1" density that the smallest possible error in the reconstruction of the analog signals result. 12. The method according to any one of claims 1 to 11, characterized in that that for optimal coding of binary sequences B with n binary digits, the #e binary In a code with code words of the same length, ones contain the binary sequences B - each written as a single-column matrix sJ - each with a k-row and n-column coding matrix £ Cj are multiplied and the result of this multiplication is output as a code word. 13. The method according to claim 12, characterized in that the matrix [C] contains only elements from the Galois field GF (p) and that # d-1 = 2e chosen arbitrarily Columns are linearly independent of ECj. 14. The method according to any one of claims 1 to 11, characterized in that that for optimal coding of binary sequences 3 with n binary digits, the <e binary ones contain the operation in a code with codewords of the same length = h (x) + is executed, where f (X) = b1Xn-1 + b2Xn-2 ... bn-2X + bnX0 g (X) = g1XK + g2XK-1 ... GK + 1X0 are polynomials with coefficients from GF (P) and the eo-coefficients b. correspond to the binary digits of the matrix and that g (X) is the generator polynomial of a error-correcting codes with the minimum distance d and that as a code word only the coefficients of the remainder polynomial r (X) are output. 15. The method according to any one of claims 12 to 14, characterized in that that on the transmission side the sequences to be coded are checked with more than e ones whether the ones appear as a bundle, denoted by a code of length n with the generator polynomial g (X) can be corrected as an error bundle in the code word, and that if this bundle can be corrected, the coefficients of the residual polynomial r (X) are sent as code word K1. 16. The method according to any one of claims 12 to 15, characterized in that that received code words [K] = r (X) are coded back into the binary sequences rBg, in that [B] C f (X) is calculated from r (X) in the same way as with known Fahlerd correction methods the error word ff (X) is calculated from the syndrome S (I). 17. The method according to any one of claims 12 to 16, characterized in that that the recipient is informed by a special character added to the code word, whether in the reconstruction of f (X) - EBS from r (X) # tK) the error message or that Method of correction method for statistically distributed errors is to be applied and the receiver uses the relevant method for decoding. 18. The method according to any one of claims 12 to 17, characterized in that that when receiving inadmissible code words they are considered code words with transmission errors be marked. 19. The method according to any one of claims 12 to 18, characterized in, that the binary sequences [B] from initially all first binary digits of the Gray code words of a line are formed, then from all second binary characters and then formed from all third, etc. binary characters of the Gray code words of a line and in this way a spreading or restructuring of single bundles is achieved will. 20. The method according to previous claims, characterized in that that on the receiving side the disrupted binary characters received by a disturbance detector are registered and that 1. disturbed sections in a line or 2. disturbed Code words in a line or 3rd disturbed lines entirely through the binary characters of the the previous line. 21. The method according to the preceding claims, characterized in that that the code words gEj - r (X) are protected by a redundant code on the transmission side and that transmission errors due to this code on the receiving side in can be eliminated in a known manner. 22. The method according to previous claims, characterized in that that a line or parts of a line which are neither by the RLC method nor by the OC process can be coded, identified by a special character and that the recipient by recognizing this special character through the relevant line replaces the previous line or at least the corresponding parts of the line be replaced by the corresponding parts of the previous line. 23. The method according to previous claims, characterized in that that at the beginning of a picture a picture synchronization character is transmitted and that the receiver determines the beginning of the line by counting the received binary characters and the reconstructed video signal the line synchronization pulses automatically adds. 24. The method according to previous claims, characterized in that that with the help of an image memory that records the code words according to the RLC coding, the modulo 2 difference of successive images is formed and that these difference binary sequences be coded again to reduce redundancy. 25. The method according to previous claims, characterized in that that only the image difference binary sequences of a line are processed further, if this has a lower "1" density or cheaper than the original line Have a "1" structure and that otherwise the original binary line sequence is further processed will. 26. The method according to previous claims, characterized in that that with the help of an image memory, which the signals after the reduction of the "1" density the modulo 2 difference of the consecutive ones represented as binary characters Images is formed and that this difference reduces redundancy in parallel by a Runlength coding or optimal mapping is coded. 27. The method according to previous claims, characterized in that that on the transmitting side the number i of ones in a sequence rB1 is determined and that different i for different i in the optimal coding method Codes with different lengths are used and that the nmpfänger the number i of the ones in a sequence EBj in addition to the transmitted non-reaundance Code word K 'is communicated, this communication only taking place when a code change has taken place. 28. The method according to previous claims, characterized in that that the runlength coding is implemented as burst runlength coding and that before A sequence of e zeros is sent to each address, with a sequence as a burst is defined by ones and zeros in which there is no sequence of consecutive zeros of length <3 occurs and that the address is the distance between consecutive Indicating bursts.