DE3029190A1 - Differential PCM system for TV line reduction code - uses transform coding and spectral coefficient prediction to reduce TV transmitted line code - Google Patents

Abstract

Bit rate reduction coding for the transmission of TV picture information over data transmission facilities at a maximum of 10 kbit/s. The picture , e.g. 100 000 spot images quantised into a 6 bit code representing 64 grey scale steps, is transmitted in ideally 1.5 seconds and at most 3 seconds. A transform code is used to convert the picture signal into a number of spectral coefficient values for transmission in a frame structure. DPCM (Differential PCM) transmission is used with a transmitted prediction for correlating picture samples. Each transmitted frame, w, of a picture, m, is derived by using actual picture spectral coefficient, y(u,v,w,m,), and previous picture (m-1) spectral corresp. coefficient, information so that prediction coefficient information is included. Consequently the difference value for each block, w, results. Further processing derives magnitude classifications, c(w,m), for each sample in the overall picture. The signal from the TV camera (1) is A/D converted (2) and stored (3) in a 16 x 16 format. Transform coding (4) and coefficient classification, DPCM prediction and quantisation (7,8,9,10,11) then precede modem (12) transmission.

Description

description

"Pseudo Moving Image Transmission System" The present invention relates to a pseudo moving image transmission system for the transmission of fast image sequences via narrowband message transmission:; channels e using transform coding in which samples of the image signal can be converted into spectral coefficients in which the image to be transmitted is divided into '' partial images is decomposed, the transformation of the samples of the image signal of the partial images is carried out in blocks and a different number of spectral coefficients per field selected and transmitted one after the other.

A moving image transmission requires the use of data compression techniques a signal rate to be transmitted for a color television picture at 25 frames per second of about 34 Mbit / s and for a picture television signal about 2 Mbit / s.

However, they are only narrowband Messaging channels with a maximum bit rate of 10 kbit / s is available, it is a moving image transmission the known type is no longer possible. The only way out is the artificial transmission of moving images, d. That is, the transmission of still images in rapid succession.

The transfer of an image from e.g. B. 100,000 pixels in 6 bits are quantized corresponding to 64 gray levels, takes a narrow band Message transmission signal with a bit rate of, for example, 10 kbit / s approximately bO seconds. However, this time is long for observations of motion processes too long. With methods of data reduction by exploiting relevance and irrelevance In the image, the sequence of images can be shortened by about an order of magnitude. A such a method is e.g. B. has become known from DE-OS 28 35 434. With this one A transformation coding is used in the method, in which the sample values of the Image signal can be converted into spectral coefficients. To do this, the Image broken down into partial images, the transformation of the sample values of the image signal Sub-images carried out in blocks and a different number for each sub-image selected by spectral coefficients and transmitted one after the other.

At the receiving end, the spectral coefficients are stored in an image memory stored and continuously an inverse transformation carried out so that the image constantly displayed and observed on a monitor during its construction phase can be. However, the frame rate achievable with this method is for many Applications still too big.

Investigations have shown that for a pseudo moving image transfer, an image sequence time between 1 and 3 seconds which is to be strived for. With shorter times, the eye is irritated because the individual images are no longer there can record, with times longer than 5 seconds, the sequence of movements is often not more recognizable.

The object of the invention is to provide the state of the art to enhance. In particular, in a system of the type mentioned at the outset, the coding of the data to be transmitted are selected so that communication channels with a maximum permissible bit rate of 10 kbit / s in each case within a maximum of 3 Seconds an image can be transferred.

The object is achieved by the invention mentioned in claim 1. It is now possible at data rates of, for example, 10 kbit / s image sequence times from 1.5 £ 3 seconds. This allows motion processes to be observed Scenes such as B. in traffic surveillance or fire observation, tracked well will. If transmission channels with a higher bit rate are available, the image sequence time further reduced or increased the resolution of the images.

Advantageous further developments and refinements of the invention are given in the subclaims. By using the cosine transform the sampling values of 16 x 16 pixels each is fast and powerful Data reduction possible. Claim 3 shows a class assignment of the blocks to the richness of detail and the movement of the partial images of a full image is adapted. In the case of calm picture scenes, this results in a picture sharpness comparable to that of the television picture achieved. In the case of strongly changing A 2-dimensional avoids image scenes Low-pass on the receiving side the occurrence of stör.nd..i block boundaries in the reconstructed Image. By varying the lower class limit, the transmission channel is always optimally exploited and so transmitted an image that reflects the characteristics of the human Seeing is largely adapted. The measures mentioned in claims 6 and 7 serve for further data reduction in the control information. Will the tax information before transmission by an error-correcting code against transmission errors protected, an image transmission via disturbed is also advantageous Transmission channels possible. Claims 9 and 10 provide for error protection advantageous developments, so that a pseudo moving image transmission with 1.5 Seconds of picture sequence time via message transmission channels with 10 kbit / s bit rate and an error rate of up to 10 2 provides satisfactory image quality.

The measure according to claim 11 has the effect that after a change of scene Details in the dormant background of the muddy scene can be displayed more quickly.

The invention will now be explained on the basis of exemplary embodiments and drawings explained in more detail. They show: FIG. 1 Fuztktionssequenz the coding FIG. 2 example a bit allocation in four classes FIG. 3 Functional sequence of decoding FIG. 4 Structure of a code word of an image FIG. 5 Example of a bit allocation for a Special class In FIG. 1 is the functional sequence of the image coding according to the invention shown. The image number is indicated with m, the block number with w, and with J k and 1 the line or. Column indices of the blocks before the transformation (in the time domain) and with u and v the row and column indices of the blocks after the transformation (in the spectral range).

From the video signal of a picture scene to be recorded above Television camera 1 are sampled values of one image at short time intervals and stored in a digital image memory 3 after analog-to-digital conversion 2. The image signals are each of a partial image consisting of 16 × 16 pixels treated as a block and the blocks successively linearly in a spectral range transformed. An image consisting of about 100,000 pixels is consequently shown in 38 pound blocks disassembled. The linear transformation converts digitized pixels x (k, 1, w, m) in spectral coefficients (y (u, v, w, m).

As a linear transformation, the cosine transformation has proven to be particularly favorable, since it can be carried out very quickly, a causes no data compression and is relatively insensitive to transmission errors is.

For effective data reduction, before the actual processing de: Blocks first made an analysis of the whole picture. For this purpose, the Spectral coefficients y (u, v, w, m) of an image are stored in an image memory 5, for which the memory 3 can also be used, since one after every data information Blockes the ladies required for this in memory 3 are no longer required.

Subsequently, a picture-to-picture difference formation of the Specific coefficients carried out and containing the resulting difference values Blocks each assigned to one of several classes.

For picture-to-picture DPCM coding, a quantization 7 with a plurality of controllable quantizers provided and a memory 6, the quantized from Difference values reconstructed image stores each for the duration of an image sequence.

The unit 11 serves, on the one hand, to determine optimal prediction coefficients p for the DPCM and, on the other hand, the determination of a classification variable c each Block.

In the unit 10, each block is classified into one the specified classes and the assignment of a class code. k to the block.

The purity 9 determines the scatter to the difference values in certain Class zones and unit 8 determined from the mean spread of the difference values of the zone modulation constants a for the quantizers of the DPCM quantization 7.

Investigations showed that movements in the sequence of images affect all prediction coefficients. of a block have the same effect. It is therefore sufficient to have a common prediction coefficient according to the equation for all spectral coefficients of a block to investigate. Here: yw (u, v, m) are the spectral coefficients of the wth block and yw (u, v, m-1) are the quantized spectral coefficients of the wth block of the picture preceding the picture m. With this prediction coefficient assigned to a block, a classification variable c (m) for the block w is now obtained according to the equation calculated and according to this value the block is assigned to one of several classes.

In FIG. 2 are, for example, four classes that are beneficial for data compression and have shown the image reconstruction on the receiving side, each in matrix form shown. After the image analysis has been completed, the fields to be transmitted are to be found in the fields Difference values y = y (u, v, w, m) - p (w, m) y (u, v, w, -m-1) one Block with the number of bits specified in the field. Since these difference values especially when using the cosine transformation very quickly with the sum (u + v) of their row and column indices decrease, the bit assignments are chosen so that for u = v = 1 the highest number of bits, for example 7, is made available. A quantization of the difference values with only one bit is not provided, since this quantization is unfavorable when reducing quantization errors over time shows in the reproduced image.

On secondary diagrams (u + v constant) there are constant bit assignments chosen. Are areas for differential values to be quantized with the same number of bits also divided into zones z, which are usually not wider than three neighboring zones Adjacent diclgonals are chosen. In the class k = 1 there are, for example, seven zones with 7, 6, 5, 4, 3 and 2 bit values. In class k = 4, which is for poorly detailed Blocks representing areas are provided, however, there are only two zones 19 and 20 with 3 and 2 bit values.

So that the difference values through the number of bits specified in the zones Can be quantized with little error - are the respective difference values a zone with a modulation factor a (z), which is determined separately for each zone will evaluate. Every modulation factor and thus every zone of the four classes a quantizer is assigned in the quantization 7 (FIG. 1), its characteristic can be set using the modulation factor. As the four classes in total, for example 20 zones, in this case 20 modulation factors are to be determined and by this 2C quantizer in the quantizer 7 to control.

A size proportional to the mean spread of the difference values in the zone a (u, m) is used. The determination of the spread of the differential values of a zone is determined, as from FIG. 1 can be seen, the unit 9, and the mean dispersion and thus the modulation factors a are determined by the unit 8.

From the determined prediction coefficients p, class key figures k and modulation factors a, control information is now formed for each image, the In this example for 384 blocks of an image from 384 prediction coefficients, 384 class indicators and 20 control factors.

In the channel coding 12, a code word is now per picture from a synchronization word, the control information and the means of the determined values of the 'expensive information quantized difference values obtained by image-to-image DPCr performed and passed the code word to the transmission channel.

In FIG. 3 shows the functional sequence of the decoding. the Image decoding is essentially the reciprocal of the coding on the transmission side, whereby the data of the control information simplify the decoding considerably, since the analysis part is omitted. The unit 21 for channel decoding recognizes the beginning of a code word at the synchronization word preceding this, separates the Control information and sends the required control variables to the DPCM decoder 22 with DPCM memory 23 to. The reconstructed spectral coefficients y are block wci- se transformed back into digital sample values x by a unit 24 and stored in a memory filed.

A frame memory 27, which has a D / A converter 28 with a Monitor 29 is connected, takes over each after complete decoding Image the digital samples during a 50 Hz image period, for example from the image memory 25.

The refresh memory is not absolutely necessary. the Viewing still images with sudden changes is, however, subjectively more pleasant as a continuous block-by-block renewal of the image during the duration of the sequence of images, In the exemplary embodiment explained above, the class boundaries were firmly specified. This is not the case with the further development of the invention according to claim 3 more the case. Starting from a given lower class limit, each at 4 times the value of the previous class limit, the further class limits set. After the classification, the number of blocks per class becomes and of the given number of bits per class, the amount of 3it to be expected for the relevant image determined.

It is now advantageous if the maximum permissible value is not reached Bit sum for an image (determined by the maximum data rate of the: 4 message transmission channel and the image sequence time is given) to lower the lower class limit and in at least to provide additional 2-bit zones for additional differential values in a class and to quantize. These additional 2-bit zones are particular in the class k = 3 (FIG. 2) favorable, since these are little moved, but also rich in detail Image parts is assigned and this detailed image part is particularly sharp be reproduced.

When exceeding the maximum permitted amount of bits for an image, such as it is the case, for example, with strong image movements or a camera panning, Advantageously, differential values with a higher sum of their indices are not processed by the quantizer, et, It turns out, however, that if difference values of the Class ka 1 (FIG. 2) with the sum of their indices u + v (10 are suppressed, themselves On the receiving side, the boundaries of the partial images in the reconstructed image are small square ones Make patterns noticeable in a disturbing way. In order to avoid these disturbances, it is advantageous to at the receiving end, the back-transformed digital sampled values x before storage in the image memory 26 (FIG. 3) through a 2-dimensional low-pass filter 25.

To identify the additional control functions, according to a further feature according to the invention, a movement index b (m) is assigned to the image, the, z. B. quantized with 6 bits, 64 different control commands with respect to the the quantizers to be set lower class limit and the number and location further difference values to be coded or not to be coded are assigned and the corresponding settings on the receiving end when decoding the DPC11 decoder causes and for movement indicators that the DPCM decoder 22 control so that z. B. Difference values of class k # 1 with sum indices u + v (10 are suppressed, the necessary low-pass characteristic of the low-pass filter 25 sets as 2-dimensional low-pass filter 25 has a planar structure that is easy to implement ranstersal filter, the horizontally and vertically the impulse response (1; 3; 4; 3; 1) / 12 has proven particularly useful when a specified value of the movement index is exceeded b is switched on.

When determining the classification parameters that are a measure of the Is the wealth of detail in each block, equation 2 became the sum of the squares of the essential differential values of a block are calculated. Since the one with the spectral colaponents y (1, 1, w, m) and y (1, 1, w, m-1) formed the difference value only the change of the mean gray value of the partial image assigned to the block does nothing, however says about the detailed content, it can be used when determining the classification parameters c (w, m) excluded. or only considered with a low rating will. In this way, the sharpness of a still moving image is improved. When determining the prediction coefficients of the blocks according to equation 1 are however, the above-mentioned spectral coefficients with the indices u. = v = 1 must also be taken into account.

A further reduction in data for the tax function can now be achieved achieve that for each class a mean preliction coefficient p (k, m) is determined.

Instead of 384 prediction coefficients for 384 blocks are in the control function then only to transmit 4 prediction coefficients for, for example, 4 classes.

When processing moving parts of the image it can be observed that there the consideration of the friction coefficient is not very effective. Hence can for further savings in principle z. B. classes 1 to 3 (FIG. 2) a prediction coefficient p t 0 and class 4 a prediction coefficient pSg1 are assigned, so that on the transmission of the p-value in the control information can be dispensed with entirely. However, it is always useful when editing the difference values of class 4 send and receive the prediction coefficient set to values (1, so that quantization errors in still parts of the picture be dismantled.

In narrowband communication channels, under certain circumstances

Bit error rates s10 2 are to be expected due to channel interference.

Because the bit amounts of the blocks in the individual classes are different are destroyed by a bit error in the information about the class membership on the receiving side The correct decoding of all follow-up of the sector coefficients is therefore particularly advantageous, the control information against transmission errors by a to protect error-correcting code, which is particularly easy to implement, if for the control information of the code words to be transmitted always a constant Bit number is provided. In order to keep the effort for the driver protection low, it is advisable to assign a BCH code of block length 127 with 85 information bits Select. The coding and decoding technology of such codes is e.g. B. in the book "Error-Correcting Codes "by W. W Peterson and E. J Weldon, The MIT Press, Cambridge, Massachusetts and London (1972). The BCH code of length 127 can handle up to 6 errors correct per block. With 128 bits it can detect up to 7 errors per block will.

A code word for the system according to the invention is, for example, according to FIG. 4 is structured as follows. The beginning of the Code word forms the synchronization word with, for example, 32 bits of a pn sequence (pseudo-noise sequence). This is easy z. B. to be realized by a feedback shift register and easy to recognize. The likelihood of their being fooled by other bits of the code word is extremely small.

The control information is then sent to the synchronization word transfer. The z. B as follows: 4 x 6 bits for the prediction coefficients 24 bits 22 x 6 bits for the modulation factors 132 bits 1 x 6 bits for the movement index 6 bit 384 x 2 bit for the class code 768 bit total 930 bit with 11 BCH blocks 935 bits of control information can be saved for each 85 information bits This increases the control information to 11 x 128 -1408 bits including the Synchronization information results in 1440 bits.

With a picture: duration of 1.5 s and a maximum permissible data rate for the transmission channel of 10 kbit / s, a maximum of 15,000 bits per image are to be transmitted. This is where the 1440 bits for the synchronization information and the control information go so that about 13,500 bits for the quantized difference values to be transmitted left over.

The system according to the invention allows a large number of variants to, iie the respective application of the system Take into account. For pictures with many structures, it is useful to provide a special class, only in zones with higher total indices (u + v-) difference values with higher Number of bits than, for example, in the class k = 3 of FIG. 2 provided, quantized can be.

An example of such a class is shown in FIG. 5 shown. Classes of this type are particularly suitable for blocks and thus for partial images, in which essentially only have larger amplitudes with higher order spectral coefficients appear.

It can also be useful to use a part in the case of severely disturbed channels the lower order differential values to be transmitted (e.g. with sum indices (etc.) + v) <4) of the blocks with moving image parts, which are therefore of classes 1 to 3 of FIG. 2, with a short-length BCE code, e.g. length 7 or 15, to protect.

In the event of severe interference in the transmission channel, errors can be bundled with high density of errors occur, which can last up to a few seconds. Error correction is then no longer possible, and neither is decoding at the receiving end the spectral coefficient would lead to a completely unusable image. In these In some cases it is advisable to keep the last decoded picture before the disturbance occurs. For this purpose, the channel status can be caused by interference during demodulation on the receiving side detecting detectors are monitored. However, the fault detection can also be performed in advantageously with the help of the secured control information on the device for error detection and correction, in which at the receiving end A- direction for filing correction, the number of correction messages is evaluated in the 11 blocks.

In normal cases, however, the system according to the invention also allows an extremely low data rate per pixel (about 0.15 bits per pixel) image sequences of moving images 1.5 seconds apart via narrowband communication channels of the type described. However, there is a digital news channel with a sufficiently large bit rate available, the system is also able to process image sequences with full motion resolution, for example 25 Hz frame rate, in known television quality.

Claims (11)

Claims 1. Pseudo moving image transmission system for ¢ bertraçmg of fast image sequences via narrowband communication channels below Use of a transformation coding in which the sample values of the Bildsig3.als are converted into spectra coefficients in which the image to be transmitted is divided into partial images is decomposed, the transformation of the sample values of the image signal of the partial beams in blocks is carried out and a different number of spectral coefficients for each partial image selected and transmitted one after the other, characterized in that first for each block (w) of an image (m) using the spectral coefficients (y (u, v, w, m)) of the current image and the corresponding spectral coefficients (y (u, y, w, m-1)) of the previous image (m-1> a prediction coefficient (p (w, m)) is determined, that for each buck (w) difference values (y (u, v, w, m) - p (w, m) Ay (u, v, w, m-1)) using the spectral coefficients Cy (u, v, w, m)) of the block, the corresponding spectral coefficients (y (u, v, w, m-1)) of the previous image (m-1) and the prediction coefficient determined for the block (p (w, m)) and with these difference values a classification variable (c (w, m)) is formed that correspond to the classification variable (cw, m)) each Block of one of several classes with a given bit pattern, given zones (z) and predetermined class code (k) is assigned that for each zone (z) the mean spread of the difference values lying in the zone is determined and the scatter determined in this way as modulation factors (a (z, m)) for one of the respective Zone assigned quantization can be determined that from the determined prediction coefficients (p), class codes (k) and modulation factors (a) form control information that for a picture from one synchronization word, the control information and the control information that is now carried out using the 1's control information with the determined values Image-to-image DPCM quantized difference values obtained by forming a code word md is transmitted and that the image decoding on the receiving side after the channel decoding while evaluating the data of the control information essentially reciprocally to the sending side Coding is carried out. 2. System according to claim 1, characterized in that the transformation a cosine transformation of the sample values of 16 x 16 pixels each is used will. 3. System according to claim 1 or 2, characterized in that for Class assignment first of all for all blocks (w) of an image (m) the classification variables (c (w, m)) can be determined that starting from a predetermined lower class limit provisional class limits for the classification size each at 4 times the value of the previous class limit it is determined that if the value falls below the maximum permissible bit amount for an image, the lower class limit is lowered and / or in at least one class additional 2-bit zones for additional differential values are provided and quantized that when the maximum permissible Hit set for an image in at least one class. Difference values of certain zones not be quantized that to characterize the determined lower class limit and the number of zones to be quantized per class, a movement index (b (m)) is assigned to the image and that the control information by the Pewegungskennzahi (b (m)) is added. 4 System according to claim 3, characterized in that the receiving side In accordance with the movement index (b (m)), a two-dimensional deep piss is also available (rt) is controlled 5. System according to any one of the preceding claims, characterized characterized that when determining the classification variables (c (w, m)) difference values, which correspond to the mean gray value of a partial image assigned to the block, not be considered or only with a low evaluation. 6. System according to one of the preceding claims, characterized in that that a mean prediction coefficient (p (k, m)) is determined for each class. 7. System according to one of the preceding claims, characterized in that that only two values are provided for the quantization of the prediction coefficient are. 8. System according to one of the preceding claims, characterized in that that the control information prior to transmission by a error correcting code is protected against transmission errors. 9. System according to claim 7, characterized in that for the control information a constant number of bits is set, the control information is divided into several blocks and every block is secured by an error-correcting code. 10. System according to claim 8, characterized in that for Fehlericherun a 127-bit BCH code is used, each containing 85 information bits contains. 11. The method according to claim 1 or 2, characterized in that for blocks in which essentially only with higher order spectral coefficients larger amplitudes occur, a separate class is provided.