GB1605025A

GB1605025A - Noise reduction in electric signals

Info

Publication number: GB1605025A
Application number: GB953777A
Authority: GB
Original assignee: British Broadcasting Corp
Current assignee: British Broadcasting Corp
Priority date: 1975-04-25
Filing date: 1977-03-07
Publication date: 1981-12-16

Description

(54) IMPKOVEMENTS RELATING TO NOISE REDUCTION IN ELECTRICAL SIGNALS (71) We, BRITISH BROADCASTING CORPORATION, a British Body Corporate, of Broadcasting House, London, WIA IAA, do hereby declare the invention for which we pray that a patent may be granted to us, and the method by which it is to be performed, to be particularly described in and by the following statement: This invention relates to a method of and apparatus for reducing the effect of noise in an electrical input signal which is obtained by scanning pictorial information, such as a television signal, and in particular is concerned with improvements in and modifications of the invention of our British Patent No. 1,515,551 (our earlier patent).

Our earlier patent describes and claims a method of and apparatus for reducing the effect of noise in an electrical input signal which is obtained by scanning pictorial information on a field-by-field basis, to provide an output signal, in which a signal derived from a preceding-field output signal is subtracted from the input signal for the current field to provide a difference signal, low-amplitude portions of the difference signal are attenuated relative to highamplitude portions thereof, and the thusattenuated signal is added to the precedingfield output signal to provide an output signal for the current field.

The attentuation is preferably achieved by a multiplier one input of which is coupled to the subtractor output and the other input of which is coupled to the output of a non-linear transfer characteristic element which is also coupled to the subtractor output. Apparatus in accordance with this and the preceding paragraph will be referred to as "apparatus of the type specified".

This invention relates to various improvements in and modifications of the invention of our earlier patent.

The invention, which is defined in the appended claims, will be described, by way of example, with reference to the drawings accompanying the provisional specification of Application No. 9537/77 in which: Figure I is a block circuit diagram of noise-reduction apparatus embodying the invention in one aspect, and based on Figure 5 of our earlier patent; Figure 2 shows the transfer characteristic of a circuit element which can be used in Figure 1; and Figure 3 is a block circuit diagram of a modification of the apparatus of Figure 1; and with reference to the drawings accompanying the provisional specification of Application No. 12972/78, in which: Figure 4 is a block circuit diagram of noise-reduction apparatus embodying the invention in one of its aspects, and is based on Figure 4; Figures 5 and 6 are block circuit diagrams of two modifications of part of the apparatus of Figure 4; Figure 7 is a block circuit diagram of the noise measurement circuit used in the apparatus; and Figure 8 is a block circuit diagram of part of noise-reduction apparatus embodying the invention in another aspect; and with reference to the accompanying drawings, in which: Figure 9 is a block circuit diagram of a circuit for detecting "global" motion and shot changes in the picture; and Figure 10 illustrates the modifications required to the circuits of Figures 5 and 6 to enable them to be used with the detector of Figure 9.

In accordance with a first aspect of the invention, where a low-pass filter is included between the output of the subtractor and the input to the non-linear transfer characteristic element, there is further included a rectifier connected between the subtractor and the low-pass filter.

Referring to the example shown in Figure 1, the input 10, subtractor 12, delay 28, multiplier 26, low-pass filter 22, function generator 24, adder 16, delays 20a and 20b, predictor 30 and output 18 are all identical to the same-numbered elements shown in Figure 5 of our earlier patent, to which reference should be made for a full description of them. Briefly, assuming that the scanned signal is an interlaced-scan television signal, that is each picture is made up of two fields, each input signal has subtracted from it in subtractor 12 the output signal for the preceding picture period as supplied by the delay device 20a to produce a difference signal representing the differences between the current and preceding picture periods.

This difference signal is passed through the attenuator 14 which is arranged to attenuate low-amplitude difference signals by a greater factor than high-amplitude difference signals, and the thus-attenuated signal is then added back in adder 16 to the output signal for the preceding picture to provide the output signal for the current picture.

Thus, low-amplitude difference signals are assumed to represent noise and are attenuated. High amplitude difference signals are assumed to represent movement and suffer less attenuation.

We have now found that improved operation results from including a rectifier 100 between the subtractor 12 and the low-pass filter 22. The filter 22 is here assumed to be a 2-dimensional transversal filter, so that the filter output which is associated with any picture point is derived from the values over an area of the picture in the neighbourhood of that point. The effect of the rectifier 100, which takes the modules of the applied signal, is to produce approximately a measure of the root mean square (RMS) value of the difference signal at the filter output.

Strictly this is true only for an equalcoefficient transversal filter with the filter input samples uncorrelated. In this way a measure of the power of the difference signal is obtained. This we have found to give subjectively improved noise reduction.

In accordance with a second aspect of the invention, the non-linear transfer characteristic element (e.g. circuit 24) is such that when the signal applied to it is above a predetermined value a constant predetermined minimum attenuation factor applies, preferably equal to unity (i.e. no attenuation).

Figure 2 shows a transfer characteristic having this feature, and plots the multiplier value of multiplier 26 against the voltage applied to the circuit 24, the latter being expressed in arbitrary units. It is seen that at a voltage of 3 units the multiplier value becomes unity (point A) and holds this value for higher input signals. This contrasts with Figure 4 of our earlier patent, where the curve approached unity asymptotically.

The shape of the curve between point A and the point B where the predetermined maximum attenuation obtains is given by the function c( 1 -x2), where x is the voltage in arbitrary units, and c is a constant greater than one. Normally c is less than 2, and a preferred value is 9/8.

It is preferred that the input voltage VA at which the curve reaches its final value should be related to the voltage VO where the curve intersects the abscissa by a factor of at least Wand preferably less than 4. As shown in Figure 2, this factor is 3.

In Figure 2 the attenuation factor K for low-amplitude signals approaching zero is taken to be 4.

In accordance with a third aspect of the invention a variable gain element is included between the low-pass filter and the nonlinear characteristic transfer element.

This is illustrated in the example of Figure 3, in which the filter 22 again takes the form of a two-dimensional averager, and where a variable-gain amplifier 102 is included between the filter 22 and the circuit 24. As used with the rectifier 100, the gain is manually adjusted so that the RMS value of the difference signal corresponding to the noise level lies at the point B where the non-linear characteristic begins to rise. The motion detector is then optimally sensitive to motion. In practice the finite spread of the RMS signal may require it to be set somewhat above or below this point, depending on the exact compromise between noise reduction and loss of moving detail.

The three modifications thus-far described can thus be applied independently to the method and apparatus of our earlier patent. They are, however, preferably used in combination, and they help to ensure that: (i) the multiplier value varies between the basal value and unity in a slow and controlled way, well covering the area of movement; in this way the picture-topicture integration is satisfactorily switched on and off; (ii) no peculiar effects occur for certain moving patterns, such as gratings, which might not activate the movement detector of our earlier patent; and (iii) moving texture is detected even when it produces a quasi-random picturedifference like noise.

With the apparatus of our earlier patent the "texture" signals would have been lost by the filter and so not have activated the movement detector. However, their presence can be detected by power measurement, because their random nature still increases the power of the picturedifference signal.

The use of the variable gain element, described above with reference to Figure 3, is of primary importance when the transfer characteristic of the non-linear element is such that below a predetermined value a constant maximum attenuation factor applies, and above this value the attenuation factor progressively reduces. The variablegain element can then be adjusted manually so that the mean RMS noise level in the signal corresponds to the said predetermined value. The system is then optimally sensitive to motion.

This is a useful feature, but in practice the noise level in a signal can vary, particularly as between shots for example, and this would require continuous monitoring and adjustment to obtain the best results. This is impracticable.

Preferably, therefore, the variable-gain element is automatically controlled by a noise monitoring circuit.

A circuit embodying this feature is illus traced in Figure 4, much of which is similar to the previous Figures and will not therefore be described again in detail.

Briefly, therefore, the circuit of Figure 4 receives an interlaced-scan television signal at an input 10 which is applied to a subtractor 12. The input signal for each picture period has subtracted from it the output signal for the preceding picture as supplied by delay device 20a to produce a difference signal representing the difference between the current and preceding picture periods. This difference signal is passed through an attenuator 14, which includes a multiplier 26 controlled by a circuit 24, and which is arranged to attenuate lowamplitude difference signals (assumed to be noise) by a greater factor than highamplitude signals (assumed to represent movement). The thus-attenuated signal is then added back in adder 16 to the output signal for the preceding picture supplied by delay device 20b to provide the output signal for the current picture.

The circuit 24 is a non-linear transfer characteristic circuit preferably having a characteristic as shown in Figure 2. That is to say, below a predetermined value a constant maximum attenuation factor applies, and above this value the attenuation factor progressively reduces, until it reaches unity at a second value which is about three times the first value. To control the circuit 24, the difference signal from subtractor 12 is applied through a rectifier 100, a low-pass filter 22 (preferably a two-dimensional transversal filter), and a variable-gain element 102 constituted by a multiplier, to the circuit 24.

A delay element 28 compensates for any delays introduced by the filter 22. It is because of this that the delay device 20b is required as well as the delay device 20a, as described in our Patent No. 1,515,551.

The variable-gain element can be adjusted, with a view to making the RMS value of the component of the difference signal corresponding to the noise level approximately equal to the predetermined value below which the maximum attenuation factor applies. As described above with reference to Figure 3, this adjustment is made manually.

In Figure 4, however, there is an additional circuit X in which the noise is measured electrically and which automatically controls the multiplier 102. The noise measurement is made in a noise measurement circuit 202 the output of which is applied to a reciprocal circuit 200, which in turn controls the multiplier 102. In this way the noise measurement can be updated once per field or per picture, and can control the multiplier 102 so that the signal level corresponding to the RMS noise level is kept at the input to the function circuit 24 to a desired fixed value, near to the said predetermined value.

The construction of the noise measurement circuit will be described below with reference to Figure 7.

The reciprocal circuit 200 produces an output signal which is inversely proportional to its input signal, and this output signal is the multiplier value. The reciprocal circuit can take the form of a read-only memory.

In Figure 4 the measurement is made on the signal prior to the multiplier 102, this being an open loop or forward feed type of control. Alternatively a closed loop or feedback type of control can be adopted as shown in the modification of Figure 5.

In Figure 5 the noise measurement circuit 202 is connected to the output of the multiplier 102. The required reference level nO, is then subtracted from the actual noise measurement in a subtractor 204, and the resultant error signal (plus or minus) is applied to a non-linear circuit 206, conveniently a read-only memory, which produces an output signal which represents the factor by which the current multiplier, stored in a store 208, must be multiplied to assume the correct value in the next field.

The output signal from circuit 206 therefore multiplies in a multiplier 210 the output of the store 208 to generate a new input for the store. The stored value, which is held for a complete field or picture, is applied to the multiplier 102.

If the noise on the preceding field or picture was n and on a particular field or picture under consideration becomes n + An, the signal values are as shown on Figure 5. The multiplier held in store 208 for the preceding field or picture will be nin so that the output of the noise measurement circuit is nJn times n + An which is nO (1 + Ann). From this nO is subtracted in subtrac tor 204 to leave nOn/n.

The non-linear circuit 206 operates on the signal n0An/n to produce an output (1 + Anln)-l [It can be seen from this that the function required of circuit 206 is that from an input signal x it must form an output signall + x/nO)~I.] This signal (1 + An/n) multiplies the value no/n stored in multiplier 210 to give n2(n + An), which is the required value for the field or picture under consideration.

Preferably the multiplier 102 is of logarithmic form so that it can handle a large dynamic range of noise levels. The control loop is then modified as shown in Figure 6.

In this case the non-linear circuit 206 produces from an input x an output -log (1 + (x/nO)) . Thus from an input nOn/n it produces an output of - log (1 + An/n) and this is added in an adder 212, replacing multiplier 210, to the value stored in store 208 to generate a new input for the store of the form log (noun + A n)).

In either of Figures 5 and 6 the multiplier 102 can if desired precede the filter 22.

In the case of either Figure 5 or 6 quantisation of the signals in the non-linear circuit 206 will cause the multiplier value to oscillate about the theoretically required value, with a peak-to-peak variation of one quantum step. This idling or hunting behaviour can be made sufficiently negligible compared with noise measurement errors by suitable choice of the quantum step. It is also advisable to limit the range of the fixed relationship contained in the circuit 206, i.e.

to limit the magnitude of its output, to guard against sudden changes due to shot changes.

The noise measurement circuit 202 will now be described in more detail. The object of the circuit is to detect the lowest value of RMS picture difference during each field, this being assumed to represent the noise level. In theory this value could be zero, but the smoothing of the spatial filter 22 ensures that the probability of this happening is very small. The RMS picture difference due to the noise alone can be considered to have a Gaussian probability distribution, with a mean and standard deviation proportional to the RMS noise input. The lowest value detected lies somewhere on the lower tail of this distribution with its own probability distribution.

Referring to Figure 7, the input 220 is connected to one input of a selector switch 222, the output of which is connected to a one-word store 224. The output of store 224, as well as being connected to the other input of the selector switch 222, is connected through a switch 226 to an accumulator 228 and also to one input of a comparator 230 the other input of which is connected to the circuit input 220. The comparator output is applied through a gate 232 to control the position of the selector switch 222.

Line pulses are applied from a terminal 236 to control switch 226 via a gate 234.

Both gates 232 and 234 receive an enabling signal from a terminal 238. When switch 226 is closed the value in store 224 is applied to the accumulator 228, the output of which is applied in turn to a normaliser circuit 240.

The output of the circuit 240 constitutes the output of the noise measurement circuit 202. A clear signal can be applied to a terminal 242 to clear the accumulator, and a set maximum signal is applied to the store 224 from a terminal 244.

In operation, comparator 230 compares the input signal (A) with the stored value (B) in store 224 and controls switch 222 so as to select the lower of the two. Thus if B is less than A, then B is simply rewritten into the store, while if A is less than B, the stored value B is replaced by the new value A. Thus the stored value represents the minimum value obtaining. The store 224 is set to a maximum value from terminal 244 at the start of each television line and at the end of the line the store content is transferred to the accumulator 228, which is cleared at the start of each field. At the end of each field the accumulated sum of the minimum values is divided by the number of lines, and this quotient represents the average minimum over all the lines. This is used as the noise measurement.

To avoid false measurements the noise measurement circuit 202 should be disabled outside the active picture area, that is unless the predominant noise has been added to the composite signal. Preferably the area of measurement inhibition should extend somewhat into the active picture area to allow for the group delay of the filter 22. An appropriate enabling signal is applied to terminal 238 and controls gates 232 and 234.

When gate 232 is disabled, switch 222 recirculates the stored value B.

We have found that there is a problem associated with the variation in noise level across the grey scale. For example, gamma correction of thermally generated source noise, as in a camera, produces many times more noise amplitude at black than at white.

On the other hand, gamma correction of signals produced by a telecine machine produces a film grain characteristic giving rise to noise which takes a peak value in the low level greys and is zero at black and white. If, as in these circumstances, the noise distribution over the grey scale is not uniform, then the setting of the variablegain element can only be a compromise.

The circuits thus-far described still suffer from this problem due to non-uniform distribution of the noise across the grey scale. The circuit of Figure 8 provides a desirable improvement in this respect.

described above with reference to Figure 3, is of primary importance when the transfer characteristic of the non-linear element is such that below a predetermined value a constant maximum attenuation factor applies, and above this value the attenuation factor progressively reduces. The variablegain element can then be adjusted manually so that the mean RMS noise level in the signal corresponds to the said predetermined value. The system is then optimally sensitive to motion.

If the noise on the preceding field or picture was n and on a particular field or picture under consideration becomes n + An, the signal values are as shown on Figure 5. The multiplier held in store 208 for the preceding field or picture will be nJn, so that the output of the noise measurement circuit is nJn times n + An which is nO (1 + Ann). From this n < , is subtracted in subtrac tor 204 to leave nOAn/n.

The non-linear circuit 206 operates on the signal nOhn/n to produce an output (1 + An/n) - [It can be seen from this that the function required of circuit 206 is that from an input signal x it must form an output signal (1 + x/nO)-t.] This signal (1 + An/n) - multiplies the value no/n stored in multiplier 210 to give no/(n + An), which is the required value for the field or picture under consideration.

In this case the non-linear circuit 206 produces from an input x an output -log (1 +(x/nO)). Thus from an input nOAn/n it produces an output of - log (1 + An/n) and this is added in an adder 212, replacing multiplier 210, to the value stored in store 208 to generate a new input for the store of the form log (noun + A n)).

In operation, comparator 230 compares the input signal (A) with the stored value (B) in store 224 and controls switch 222 so as to select the lower of the two. Thus if B is less than A, then B is simply rewritten into the store, while

In Figure 8, a level selector circuit 248 is connected to receive the preceding-field signal from the delay device 20a of Figure 4 (and from the predictor if present) and is connected to control a range switch 250 and a combining or OR circuit 252. The level selector circuit 248 determines into which of a number of ranges across the grey scale the amplitude of the signal currently falls. It then controls the range switch 250 which is connected to the output of the multiplier 102 so as to apply the signal from multiplier 102 to an appropriate one of three circuits Xl, X2 and Xj, each of which can be similar to the corresponding circuit of Figure 5 or Figure 6, and which correspond to respective ones of the amplitude ranges. It will be appreciated that three circuits are shown only by way of illustration; the number required may differ from three. In one example four ranges are used. The selector circuit 252 synchronously applies to the multiplier 102 the output of the circuit X currently in use.

The selector 248 operates on the luminance component of the television signal and compares this with predetermined decision levels defining the ranges.

Precautions should be taken to guard against false measurement. First, the processed signal may excurse into a grey range for too short a time for the spatial filter 22 to give a valid measurement. If this excursion is into an area of higher noise level than the surrounding area, the lower, false measurement will be retained and cause an incorrect setting. To overcome this, the measurement enable signal (at terminal 238 in Figure 7) is derived by shortening the range-enable periods (defined in selector 250) at their beginning and end by the group delay of the spatial filter 22.

Secondly, the processed signal may never excurse into a grey range during one or more lines. In this case no valid measurement will be transferred to the appropriate accumulator, and so the reference value is substituted. This dilutes the valid measurements on other lines and so with the open-loop control of Figure 4 the multiplier setting is incorrect. However, with the feedback arrangements of Figures 5 and 6 it merely delays the time taken to adapt.

Hence these arrangements are preferable to that of Figure 4. If the grey range is absent from the whole picture, and no valid measurements result, then the multiplier value is frozen at the value obtaining for the last scan on which a valid measurement was achieved.

If, in the arrangement of Figure 8, the circuits of Figures 5 and 6 are used, and the spatial filter 22 precedes the multiplier 102, as shown, and all settings are correct, there will still be an undesirable effect at highcontrast transitions if the noise varies markedly over the grey scale. This occurs because the spatial filter 22 smoothes the transition of the noise signal, which is then multiplied by a value having a sudden transition due to the fast grey-range selection. This causes an inflection in the signal applied to the function circuit 24 which results in a failure to noise-reduce on the lower-noise side of the transition. If the multiplier precedes the spatial filter 22, no such effect occurs because both multiplier and multiplicand can vary rapidly at the input to the spatial filter. This arrangement is therefore preferable on theoretical grounds.

It is assumed above that when the picture contains moving information there is always some part of the picture in each grey range or segment where a valid measurement of noise can be made, i.e. where no picturedifference signals due to movement are created. In practice we have found that such an assumption is invalid for most moving pictures. The result is that when the picture moves the measured noise level increases, and the resulting adaptation to the new measurement causes low-level texture to be smeared.

We have pursued two approaches in solving this problem. One has been to detect the presence of motion over the entire picture (globally), as opposed to locally, and inhibit 'upwards adaptation' if this occurs.

The other approach has been to assume that the noise level changes only at a shot change and therefore to inhibit 'upwards adaptation' at all other times. In this context 'upwards adaptation' means the adaptation to a higher level of noise, i.e. reducing the control input signal to the multiplier 102.

Downwards adaptation, on the other hand, is always permissible and cannot be caused by movement but only by measurement errors. Adaptation in this direction only is conveniently referred to as 'ratcheting'. This carries the danger that a spuriously low measurement occurring on only one field will lock the adaptation into a state from which it cannot escape until the next set of events which allows upwards adaptation.

In practice three modes of operation can be used, namely: 1. Two-way (upwards and downwards) adaptation when no global motion; otherwise ratcheting.

2. Two-way adaptation for a number of fields after global motion has stopped; otherwise ratcheting.

3. Two-way adaptation for a number of fields after a shot change, but waiting until global motion has stopped; otherwise ratcheting.

The first mode of operation is critically dependent on the threshold of the "global motion" decision. For small amounts of motion, therefore, it allows upwards adaptation and does not cure the problem, i.e.

small amounts of localised motion are smeared. However, global panning shots are adequately dealt with.

The second mode of operation is not so susceptible because after motion there is a good chance of a stationary scene occurring when the right setting will be obtained by ratcheting. Thereafter the setting is not disturbed until after the next sequence of global motion. The mode will fail, however, if the global motion is such as to oscillate about the threshold for detection so that a stationary scene never occurs.

The third mode of operation is safest in that the setting is least disturbed but carries the danger of ratcheting to too low a level if the intervals between shot changes are very long. It is also defeated by "lap dissolves" or fades between shots.

The detection of global motion and shot changes may be carried out by the apparatus of Figure 9, which accepts a signal from the spatial filter 22 or from the rectifier 100 and integrates this in an integrator 300 over the whole of the active field period. This signal thus represents the fieid-integral of the modulus of the picture-difference signal.

The value so obtained is clocked into a seven stage shift register 302 clocked at field rate. In this way eight successive measurements are available at once. Global motion is sensed by summing in an adder 304 all the measurements except the central pair and comparing this sum in a comparator 306 with a suitable threshold value applied at an input 308. A shot change is sensed by comparing in a comparator 310 the sum from adder 304 with a value derived from the central pair of measurements. The machanism is roughly to compare the actual value of the central pair with a prediction based on the neighbouring values derived by means of an adder 312, multiplier 314 and constant subtractor 316 connected as shown.

The signals so obtained are applied to the measurement circuits Xl, X2, X3 described x above with reference to Figure 8 through suitable interface circuits to realise the options described earlier. The measurement circuits are slightly modified so as to permit ratcheting. Figure 10 shows this modification a plied to the circuit of Figure 6. The sign of the output of subtractor 204 is tested (shown in Figure 10 by a comparator 320 acting on its inputs) and the comparator output overrides the output of the nonlinear law circuit 206 by substituting in a switch 322 the value zero if the subtractor output is positive. Thus only negative inputs to circuit 206, corresponding to a decrease in noise level, have any effect. This ratcheting action can be enabled through a gate 324 by a signal applied to a ratchet enable terminal 326.

For option (1) mentioned above the global motion signal G of Figure 9 is applied directly to this terminal 326. For option (2) the termination of the G signal initiates a counter, clocked at field rate, which is used to produce a pulse lasting for several field periods and which, when inverted, is applied to the ratchet enable terminal 326. The count is also made conditional upon the existence of sufficient valid measurements, i.e. a minimum specified number of lines are required to produce valid measurements in any field before the operation is enabled.

For option (3) the counter is initiated by the shot change signal S of Figure 9 with the count conditional upon the global motion signal G.

WHAT WE CLAIM IS: 1. Apparatus of the type specified, including a low-pass filter connected between the subtractor and the non-linear element, and rectifying means between the subtractor and the low-pass filter.

2. Apparatus for reducing the effect of noise in an electrical signal which is obtained by scanning pictorial information on a field-by-field basis, the apparatus comprising input and output terminals, a delay device coupled to the output terminal for providing a delay time of substantially one scan, a subtractor coupled to the input terminal and to the output of the delay device for providing a difference signal, and an attenuation circuit operative to provide a varying degree of attenuation in dependence upon said difference signal; the attenuation circuit comprising rectifying means coupled to the output of the subtractor, a low-pass filter coupled to the output of the rectifying means, a non-linear transfer characteristic element coupled to the output of the low-pass filter, and multiplication and adding means coupled to the output of the non-linear transfer characteristic element for attenuating differences between the input signal during the current scan and the output signal during a preceding scan to provide an output signal which is applied to the output terminal.

3. Apparatus according to claim 1 or 2, wherein the non-linear element has a characteristic such that when the signal applied to it is above a predetermined value a constant predetermined minimum attenuation factor applies.

4. Apparatus of the type specified, wherein the non-linear element has a characteristic such that when the signal applied to it is above a predetermined value a constant predetermined minimum attenuation factor applies.

5. Apparatus for reducing the effect of

**WARNING** end of DESC field may overlap start of CLMS **.

Claims

**WARNING** start of CLMS field may overlap end of DESC **. dependent on the threshold of the "global motion" decision. For small amounts of motion, therefore, it allows upwards adaptation and does not cure the problem, i.e. small amounts of localised motion are smeared. However, global panning shots are adequately dealt with. The second mode of operation is not so susceptible because after motion there is a good chance of a stationary scene occurring when the right setting will be obtained by ratcheting. Thereafter the setting is not disturbed until after the next sequence of global motion. The mode will fail, however, if the global motion is such as to oscillate about the threshold for detection so that a stationary scene never occurs. The third mode of operation is safest in that the setting is least disturbed but carries the danger of ratcheting to too low a level if the intervals between shot changes are very long. It is also defeated by "lap dissolves" or fades between shots. The detection of global motion and shot changes may be carried out by the apparatus of Figure 9, which accepts a signal from the spatial filter 22 or from the rectifier 100 and integrates this in an integrator 300 over the whole of the active field period. This signal thus represents the fieid-integral of the modulus of the picture-difference signal. The value so obtained is clocked into a seven stage shift register 302 clocked at field rate. In this way eight successive measurements are available at once. Global motion is sensed by summing in an adder 304 all the measurements except the central pair and comparing this sum in a comparator 306 with a suitable threshold value applied at an input 308. A shot change is sensed by comparing in a comparator 310 the sum from adder 304 with a value derived from the central pair of measurements. The machanism is roughly to compare the actual value of the central pair with a prediction based on the neighbouring values derived by means of an adder 312, multiplier 314 and constant subtractor 316 connected as shown. The signals so obtained are applied to the measurement circuits Xl, X2, X3 described x above with reference to Figure 8 through suitable interface circuits to realise the options described earlier. The measurement circuits are slightly modified so as to permit ratcheting. Figure 10 shows this modification a plied to the circuit of Figure 6. The sign of the output of subtractor 204 is tested (shown in Figure 10 by a comparator 320 acting on its inputs) and the comparator output overrides the output of the nonlinear law circuit 206 by substituting in a switch 322 the value zero if the subtractor output is positive. Thus only negative inputs to circuit 206, corresponding to a decrease in noise level, have any effect. This ratcheting action can be enabled through a gate 324 by a signal applied to a ratchet enable terminal 326. For option (1) mentioned above the global motion signal G of Figure 9 is applied directly to this terminal 326. For option (2) the termination of the G signal initiates a counter, clocked at field rate, which is used to produce a pulse lasting for several field periods and which, when inverted, is applied to the ratchet enable terminal 326. The count is also made conditional upon the existence of sufficient valid measurements, i.e. a minimum specified number of lines are required to produce valid measurements in any field before the operation is enabled. For option (3) the counter is initiated by the shot change signal S of Figure 9 with the count conditional upon the global motion signal G. WHAT WE CLAIM IS:

1. Apparatus of the type specified, including a low-pass filter connected between the subtractor and the non-linear element, and rectifying means between the subtractor and the low-pass filter.

5. Apparatus for reducing the effect of

noise in an electrical signal which is obtained by scanning pictorial information on a field-by-field basis, the apparatus comprising input and output terminals, a delay device coupled to the output terminal for providing a delay time of substantially one scan, a subtractor coupled to the input terminal and to the output of the delay device for providing a difference signal, and an attenuation circuit operative to provide a varying degree of attenuation in dependence upon said difference signal; the attenuation circuit comprising a non-linear transfer characteristic element coupled to the output of the subtractor, and multiplication and adding means coupled to the output of the non-linear transfer characteristic element for attenuating differences between the input signal during the current scan and the output signal during a preceding scan to provide an output signal which is applied to the output terminal; and wherein the non-linear element has a characteristic such that when the signal applied to it is above a predetermined value a constant predetermined minimum attenuation factor applies.

6. Apparatus according to claim 3, 4 or 5, wherein the predetermined minimum attentuation factor is unity.

7. Apparatus according to claim 6, wherein below a second predetermined value a constant maximum attenuation factor applies, and between the first and second predetermined values the attenuation factor varies substantially in accordance with the function: c (1 - x 2) where x is the voltage in arbitrary units, and c is a constant greater than one.

8. Apparatus according to claim 7, wherein c is less than two.

9. Apparatus according to claim 7, wherein c is substantially 9/8.

10. Apparatus according to any of claims 6 to 9, wherein below a second predetermined value a constant maximum attenuation factor applies, and between the first and second predetermined values the attenuation factor follows a curve, which, if continued to a point of infinite attenuation (zero amplification), gives the corresponding voltage V0 which is related to the voltage VA at the first predetermined value by a factor of at least V.

11. Apparatus according to claim 10, wherein the said factor is less than 4.

12. Apparatus according to any preceding claim, including a variable-gain element connected between the subtractor and the non-linear element.

13. Apparatus of the type specified, including a variable-gain element connected between the subtractor and the non-linear element.

14. Apparatus for reducing the effect of noise in an electrical signal which is obtained by scanning pictorial information on a field-by-field basis, the apparatus comprising input and output terminals, a delay device coupled to the output terminal for providing a delay time of substantially one scan, a subtractor coupled to the input terminal and to the output of the delay device for providing a difference-signal, and an attenuation circuit operative to provide a varying degree of attenuation in dependence upon said difference signal; the attenuation circuit comprising a variable-gain element coupled to the output of the subtractor, a non-linear transfer characteristic element coupled to the output of the variable-gain element, and multiplication and adding means coupled to the output of the non-linear transfer characteristic element for attenuating differences between the input signal during the current scan and the output signal during a preceding scan to provide an output signal which is applied to the output terminal.

15. Apparatus according to claim 12, 13 or 14, wherein the variable-gain element is manually adjustable.

16. Apparatus according to claim 12, 13 or 14, including electrical means responsive to a signal in the apparatus automatically to vary the value of the gain of the variable gain element.

17. Apparatus according to claim 16, wherein the electrical means comprises noise-measurement means for measuring the noise in the difference signal and for adjusting the variable-gain element in response thereto.

18. Apparatus according to claim 17, wherein the noise-measurement means comprises means for measuring the lowest value of the difference signal in defined periods and for averaging these lowest values.

19. Apparatus according to claim 17, including a plurality of separate noisemeasurement means for the variable-gain element, and means for selectively using one of the noise-measurement means in dependence upon the magnitude of the input signal to the apparatus.

20. Apparatus according to claim 16, wherein the electrical means comprises a plurality of separate controls for the variable-gain element and means for selectively using one of the controls in dependence upon the magnitude of the input signals to the apparatus.

21. Apparatus according to claim 19 or 20, wherein there are at least three controls associated with respective ranges of the amplitude of the input signal.

22. Apparatus according to claim 19, 20 or 21, wherein the electrical means is selectively enabled in dependence upon the presence of motion over the entire scanned picture.

23. Apparatus according to claim 19, 20, 21 or 22, wherein the electrical means is selectively enabled in dependence upon the presence of a change in the total nature of the scanned picture.

24. Apparatus according to claim 22 or 23, wherein the electrical means responds differently to apparent increases and decreases respectively in the noise level.

25. Apparatus constructed and arranged substantially as herein described with reference to Figures 1 to 3.

26. Apparatus constructed and arranged substantially as herein described with reference to Figures 4 to 8.

27. Apparatus constructed and arranged substantially as herein described with reference to Figures 9 and 10.