DE2402513C2 - Method for compressing the message flow of binary-coded video signals - Google Patents

Description

4 ^r >

The invention relates to a method for compressing the message flow of binary-coded video signals, in which the analog video signals generated by an image recording device are sampled, quantized and binary-coded transmitted as a code word of length η vt on the transmission side and the sampled values after decoding the transmitted code in an image display unit on the receiving side Brightness signals are converted back, according to the preamble of the claim. «

The transmission of analog video signals requires a bandwidth of approx. 5 MHz. For portrait image transfer with video telephony one is content with a quality that a bandwidth of 1 MHz for the analog video signals required

In the case of digital message transmission, there is hardly any deterioration in the image quality if 7 to 8 bits are expended per sample. Taking into account the sampling theorem, this means bit rates

70 bit / s for television quality and

14 bit / s with portrait video quality

A reduction in the message flow for image transmissions is therefore for economic reasons desirable.

In order to reduce the flow of messages, pulse-code modulation is used instead of transmission (PCM transmission) a transmission using differential pulse code modulation (DPCM). It is enough then approx. 3 bits per sample (L Stenger, "A method for effective intraframe DPCM coding", Communication from the FTZ of the Deutsche Bundespost, Darmstadt [1971]). Such DPCM procedures however, have the disadvantage that transmission errors due to error propagation are too clearly visible Image disturbances in Gestak lead to lighter or darker horizontal stripes.

In order to reduce the influences of transmission errors, DPCM coders are therefore used in some proposals used, which each contain an "integrator with a leak" at the transmitter and receiver.

This measure reduces the influence of transmission errors in that the image disturbances are no longer reflected continue to the end of the line, but are degraded in the course of ten to thirty pixels. However, a more or less clearly visible periodic stripe structure arises in the image area parallel to the image contours.

A completely redundancy-free message flow can only be obtained through optimal coding (BSTJ, J. 27 [1948], pp. 379-423 and pp. 623-656). However, this coding is technically very complex and sets an exact one Knowledge of the source statistics and a stationary signal source.

The known coding methods are almost unusable, particularly when transmitting video signals over relatively severely disturbed transmission channels with error rates of up to 10 ^-3 , as occur, for example, in local network lines.

A system for bandwidth reduction has become known through the American patent specification 30 71 727, whereby a video signal for a PCM transmission is processed in two ways. It will distinguished between a slow mode and a fast mode. In slow mode, the full amplitude of the signal sample with high resolution linear quantization and less than transmitted at normal sampling frequency. The fast mode includes nonlinear differential quantization and normal sampling rate. In the slow mode, for example, every second sample is converted into an 8-bit word while in fast mode the difference between two consecutive samples is encoded by a 4-bit word. Both modes are identified by flag bits.

The invention is based on the object of a method for compressing the message flow of the Specify the type mentioned at the beginning, in which the disadvantages mentioned, in particular the error propagation as far as possible when transmitting video signals on transmission channels with relatively strong interference can be avoided and in which, despite the high data flow reduction, the high image sharpness of the The original image is almost preserved

The solution takes place with the means specified in the claim.

The PCM double mode system according to the invention has the advantage that the error propagation is very low and transmission errors are limited to 2 to 3 sample values as a result for the viewer

significantly less disruptive than with known DPCM processes. Furthermore, an effective subsequent error correction can be achieved by means of comparatively simple additional measures. With an undisturbed Obertragungskanal the achievable ϊ image quality for codes which have an output data rate of four or more bits per pixel is approximately equivalent to a DPCM coding the same bit rate, at relatively high noise transmission channels with error rates up ru 10- ^3, the Biidqualität is much better in

The invention will now be based on exemplary embodiments explained in more detail It shows

F i g. 1 analog video signal with samples,

F i g. 2 code word rich in detail image parts,

F i g. 3 Code word for low-detail image parts, ι s

F i g. 4 block diagram of an encoder and decoder for carrying out the method according to the invention,

FIG. 5 is a block diagram of an encoder with a device for the analog derivation of the decision criterion according to the invention.

The principle of the coding process is to use the Saving of bits of the data to be transmitted; Codewords an exchange between contrasting and two-dimensional Resolution in the displayed image. In image areas with little detail, a high contrast resolution, however, a reduced sampling frequency is used, whereas a high sampling frequency is used in detailed image parts and decreased contrast resolution.

The sampling theorem dictates that the Abtastfre- at least twice as high as the highest occurring in the signal frequency must be chosen frequency. For 5 MHz broadband television signals, a sampling rate of 10 MHz is required. However, this high sampling frequency does not have to have this J5 maximum rate over the entire image area. It only has to be ensured that the sampling rate is sufficient for the respective highest frequency contained in the video signal.

Experience shows that reducing the transmission bandwidth by half does not cause any perceptible impairment of the image in areas with little detail. In such image areas, a contrast resolution of 128 quantization levels is sufficient to avoid contour effects. a

On the other hand, there is a reduction in image areas with a high level of detail and large differences in contrast the number of stages to sixteen or eight quantization stages are permitted.

However, the full sampling frequency must be used in detailed image parts.

Two PCM modes are used here, depending on the low-detail or high-detail image area. The currently used mode of the transmitter is sent to the receiver with the corresponding code word communicated.

1 shows an analog signal VS with four sample values a \ to at as a function of time t . The sample values a \ to äs represent an image part rich in detail, the sample values a? to & t *><> a low-detail part of the image.

The following describes an example of a coder that has a data rate of 5 bits per sample According to the proposal, two 5-bit words are replaced by a 10-bit word f> 5

ΙΟ-bit words allow 1024 different code words to agree. In this example, 900 words are used to transmit the sample values detailed parts of the picture are used, the remaining 124 are used used for image areas with little detail.

In the detailed image areas, each of the two samples is quantized in 30 level steps. Goes if you start from a 246-step reference signal, the level steps 0, 1, 2, 3, ... 30 correspond to the Representative values 4, 12, 20, 28 to 244. Each representative value of the 246-step reference signal comprises an amplitude range of 8 steps. the The step height is therefore 3.2% of the dynamic range.

In the image areas with little detail, only every second sample is transmitted. The number of stages is 124, the step height is two steps from the 246 step reference signal. The lowest Amplitude is 0, the highest has 246 levels.

A particularly simple and advantageous method is to use the 10-bit transmission code accordingly to form the following scheme:

In Fig. 2, a transmission code word CW \ for detailed image parts is shown schematically. The digits acr valencies 2 ¹ to 2 ⁷ v / earth d:; rt are simply referred to as b to h . The level levels 0 to 29 are represented as 5-bit code words. The code words 11110 and 11111 remain unused in this case. Both 5-bit code words are then combined according to Table 1 to form a 10-bit transmission code word. The first 5-bit word occupies places 1 to 5 in order of decreasing significance, the second 5-bit word occupies places 6 to 10, also in order of significance fall. Table 1 shows this assignment.

Tabel

Places I 2

Value 2 ⁷ 2 ^ft 2 ⁵ 2 ⁴ 2 ³ 2 ⁷ 2 * 2 ⁵ 2 ⁴ 2 ^!

5-bit word A of the 5-bit word B of the
Sample a] sample ai

In this case, the recipient completes the digits not transmitted in such a way that ^{the digits 1.0 and 0 are in the digits 2, 2} , 2 'and 2 °.

In Fig. 3, the transmission code words CWi and CW ₃ for low-detail image areas are shown schematically.

The level steps 0, 2, 4, 6 ... 126 to be in accordance with the 10-bit transmission codeword CW ₂ Table 2 represents (7-bit word Cz. Of the sample B. a _4).

Table 2

Places

3 456789 10

Value 1 1 1 I 2 ² 2 'I " ¹ 2 * 2 ^J 2 ¹

Places 1 to 4 are permanently assigned 1, places 5 to 10 contain the values 2 ¹ to 2 * of the sample value α *.

The level steps 128, 130, 132 ... to 246 are determined by the 10-bit transmission code word ΓΗΊ according to Table 3 (7-bit word Dv .. B. of the sample a ₄ ).

Table 3

Pliit / c

I 2 .1 \ 5 6 7 8 ') Cl

Value 2 " T 2 ⁴ 2 '2 ^:

I 2 ¹

The pattern 1111 in positions 6 to 9 means that the steep 2 'has the value 1.

The assignment rules of the exemplary embodiment are always unambiguous if it is guaranteed that the input S-bit PCM code word has only 246 levels. The bit sequence 1111 at positions I to 4 or 6 check whether the signal exceeds a specified positive or negative threshold. The determined state is valid for two sampled values. For this reason, the following I-bit register 8, which contains the information about the decision, is only clocked at half the sampling frequency fi = 0.5 · f \.

The code converter 9, to which the undelayed signal of the sample a \ is fed at the input, forms code words of the type CW, (FIG. 3a) and CW, (FIG. 3b). It codes every second sample and forms the output code words intended for this purpose. In accordance with the discussed example, one of 120 code words is formed, each of which corresponds to one of the 120 amplitude levels of the level 2/256.

The undelayed signals of the sample u \ and the simply delayed signals of the sample a? fed. This code converter t0 encodes each sample. The two 8-bit output

10-bit word belongs to the image area with little detail. The 1111 group is missing. so this indicates that the transmitted lOBit code word contains two samples from the detailed image part.

In detailed parts of the picture only every second one is Sample sent. The omitted sample values, however, are derived from the sample values at the receiving end the neighboring pixels are interpolated and reinserted at the locations of the missing samples.

The encoder can basically be implemented in many versions. The following are two arrangements to the examples mentioned are described, of which the first of 8-bit PCM signals, the second is based on analog signals.

F i g. 4 shows an encoder for 8-bit PCM signals in the upper half and one in the lower half Decoder for the recovery of PCM signals coded at the transmission end.

The 8-bit PCM video signal is initially taken from a 8-bit shift register 1 buffered. The following limiter 2 reduces the dynamic range in in this example from 256 to 246 quantization levels. This means that of the total of 256 gray levels in a television picture, the last 10 light gray levels be suppressed. Of course, the analog video signal can also be dimensioned so that after the Analog / digital conversion can only occur a maximum of 246 quantization levels, whereby the downstream Limiter can be omitted.

The shift registers 3 and 4 that now follow delay the limited 8-bit signal by a sampling interval of duration I IU, 1 being the usual sampling frequency.

Then the undelayed signal of the sample a · and the doubly delayed signal of the sample a _{f are added} in the interpolator 5 and multiplied by a factor of 0.5. This procedure corresponds to a mean value formation between the samples a. and 3s. This mean value formed in this way is now derived from the simply delayed signal a? in an adder 6. one input of which is inverted. subtracted. The difference signal is used as a measure of the richness of detail.

If the differential signal is small, the fine one becomes Quantization with 120 levels selected, otherwise the 30-level coding.

The decision about the working mode to be selected is made in welding level 7. For example, it consists of two comparators. each quantized, each step height being 8/256. The binary values of these samples are combined to form a code word of the form CW \ (FIG. 2).

The effort for the code converter 10, which combines the two binary-coded samples c a \ and a> to form a code word, is very low, since this combination involves a simple bit position shift.

A switch II controlled τ jm I-bit register 8 selects the outputs of the provided code converters 9 and 10 depending on the operating mode. The code words delivered to the output of the encoder are sent with the sampling frequency h = 0.5 · f \ .

The decoder shown in FIG. 4, lower half accepts the transmitted code words in a storage register, for example a shift register 20 and feeds them to the code converter 21. This converts the code words according to their meaning into 8-bit PCM codewords. The code words are always due to the presence or absence of the 1111 bit pattern uniquely decodable. In addition, the code converter 21 also controls the mode selection switch 22.

For the interpolation mode for processing the finely quantized sample values, for example a *. switch position q. the position ρ of the mode selection switch 22 is set for processing the coarse quantized samples.

The code converter 21 has three outputs, namely r. s and ι. The first 5 bits of a code word can be removed from output r, the remaining 5 bits of a code word from output s. The outputs r and s of the code converter 1 ″ can be connected to a first contact ρ of the mode selection switch 23 by means of the switch 23.

The output t of the code converter 21 is the code words on the one hand to a first input of an interpolator 24, on the other hand to a changeover contact u of a changeover switch 25. The other changeover contact ν of the changeover switch 25 is connected to the output of the interpolator 24, which is a Provides a factor of 1/2 reduced sum signal of its input signals. The second input of the interpolator is fed by a shift register 26, the content of which is taken from the common changeover contact of the mode selection switch 22. This changeover contact of the switch 22 is also connected to a shift register 27, the δ-bit PCM code words of which are the 8-bit PCM code words at its output

at the input of the encoder and correspond to further processing, for example via a Digital / analog converter that can be converted back into video signals for subsequent image reproduction.

The changeover switches 23 and 25 are both switched at the rate of the sampling frequency fj = / i / 2. The shift register 26 is also clocked with the clock / j, but the shift register 27 with the original sampling clock f \. These clock pulses are generated on the transmit side and on the receive side by control units, not shown here to improve clarity, the mode of operation of which is not the subject of the invention and is therefore not described in more detail, especially since these are conventional control devices used in coding and decoding technology. Therefore, it is limited here only to the control function of the control unit.

The interpolator 24 calculates the omitted sample values preceding the finely quantized sample value by averaging the neighboring sample values, for example a \ in FIG. 1 from ai and a <.

The output signal of the mode selection switch 22 is temporarily stored in the 8-bit shift register 27, in order to eliminate the switching interference of switches 22, 23 and 25. It is used like the shift register 1 and 20 of the signal enhancements only.

F i g. 5 shows an analog version of the signal interpolation on the transmit side. In this version, the decision about the working mode to be selected is derived from the analog video signal. For this purpose, the signal is first delayed with two delay elements 30 and 31, each by a sampling interval of duration \ / f.

The input of the delay element 30 and the output of the delay element 31 are each connected to one of the two Inputs of an interpolator 32 connected, which half the sum signal to the first input of a differential amplifier The output of the delay element 30 is connected to the other input of the differential amplifier tied together.

The difference signal arrives at a symmetrical one

Threshold circuit 34, which only when a predetermined positive or negative threshold is exceeded Signal of one kind, for example a 1

"> corresponds to a subsequent I -bit register 8.

The switched at the output of the delay element 31

Analog / digital converter 35 feeds the already shown in FIG. 3 known code converters 9 and 10, their function, as well as that of the I-bit register 8 and des

The shift register 11 is the same as in FIG. 4 are.

Since the function of the encoder in FIG. 5 that of the one shown in FIG. 4, a further functional description of FIG. 4 is superfluous here. 5.
To avoid high-frequency image structures

i> low amplitude under certain circumstances during coding will be lost if the difference between the interpolated value and the exact value is less than is the mode switching threshold, a pre-emphasis becomes upstream of the encoder and the receiver

> n a deemphasis that compensates for this pre-emphasis downstream.

It is advisable to select the pre-emphasis so that the spectral components of the video signals in the vicinity of the upper limit frequency are increased by 6 to 9 dB

2Ί become. It is particularly cheap as a pre-emphasis to use phase-linear circuits, so that a symmetrical one in the event of steep jumps in the video signal

Overshoot is achieved.

To improve the sharpness of the picture,

She often uses "aperture equalizers" in cameras already represent the desired pre-emphasis. They are also known as "cosine" equalizers. For the method according to the invention, only the setting of the camera aperture equalizer has to be changed in this way.

3 ") change that the described increase in the high-frequency signal components takes place in the desired manner.

The receiver-side de-emphasis can be done with very little effort, for example by means of a low-pass filter with additional de-emphasis properties.

For this purpose 3 sheets of drawings

Claims (1)

Claim: Method for compressing the message flow of binary-coded video signals, in which on the transmitting side the analog video signals generated by an image recording device are sampled, quantized and transmitted in binary coded form as code words of length η and, on the receiving side, the sampled values are converted back into brightness signals in a picture display unit after decoding the transmitted code words , whereby the richness of detail of the image part to be transmitted is determined on the transmission side by comparing neighboring sampled values and a criterion is derived therefrom, whereby according to the criterion for detailed ι. Image parts a first sampling frequency is used and a coarse quantization is used and a second sampling frequency is used for low-detail image parts and a fine quantization is applied, so that only one sample and one characteristic derived from the criterion can be transmitted in a code word of length η and with the receiving end the finely quantized sample is converted back with the second sampling frequency and missing values lying between two transmitted finely quantized samples are derived by interpolation of neighboring transmitted finely quantized samples, characterized in that on the transmitting side before the comparison of neighboring samples, the amplitudes in the range of the high Videofrequjnzen raised and on the receiving side after decoding zi! ^r compensation of the frequency increase on the transmitter side and that the rough quantization is linear r> on the transmitter side in such a way that at least two adjacent sample values can be transmitted in a single code word of length η and that on the receiver side, if the identifier is missing, the two roughly quantized sample values are converted back with the first sample frequency .