WO1996009724A1

WO1996009724A1 - Video signal processing

Info

Publication number: WO1996009724A1
Application number: PCT/GB1995/002261
Authority: WO
Inventors: Graham Alexander Thomas
Original assignee: Snell & Wilcox Limited; British Broadcasting Corporation
Priority date: 1994-09-22
Filing date: 1995-09-22
Publication date: 1996-03-28
Also published as: GB2293514B; AU3529195A; GB9419137D0; GB2293514A

Abstract

Distortions in the coding or transmission of chrominance signals, such as caused by band-limiting, can cause failure of constant luminance, particularly since the signals are subjected to a non-linear gamma correction pre-processing operation. To reduce this problem, the coder includes circuitry (7, 8, 9) simulating a decoder to produce a perceived luminance value (10) which is compared with a true luminance value (3) calculated from the originating colour component signals. A comparison is made and a correction derived which is added to the transmitted luminance.

Description

VIDEO SIGNAL PROCESSING

This invention relates to the processing of video signals to provide a luminance signal which is compensated for failure of constant luminance caused by distortions in the coding or transmission of the chrominance signals. It is common practice to represent a colour video signal as a luminance component, Y, and two colour difference (or chrominance) signals, often termed U and V. One reason for representing a video signal in this way is that it is possible to restrict the bandwidth of the chrominance signals compared to that of the luminance signal with minimal subjective impairment to the image, since the eye is less sensitive to chrominance detail than to luminance detail.

However, in many such coding systems, including PAL, NTSC, and CCIR Recommendation 601 , the so-called luminance component does not precisely represent the true perceived brightness of the image. There are two reasons for this; firstly, the red, green and blue colour signals from the camera are "gamma-corrected" before the matrix that forms Y', U' and V (primes indicate gamma-corrected signals). This consists of applying a non-linear transfer function to the signals to pre-correct them for the non-linearity of standard cathode-ray tube (CRT) displays. Since the Y signal is formed from a combination of these non-linear signals, it cannot represent true brightness. Secondly, the coefficients that define the relative contributions to Y of red, green and blue do not generally have the correct values to make Y represent perceived brightness.

The effect of the Y signal not representing true brightness is that some information relating to the true brightness of the image travels through the chrominance signals. If all three signals were received intact this would not matter, as the colour decoder reverses the processes described above in the correct order, so that the RGB signals are perfectly recovered and the perceived luminance is correct. However, if only the luminance signal is received, for example on a monochrome receiver, the perceived luminance will be incorrect; this effect is known as 'failure of constant luminance'. In a colour receiver, if the chrominance signals are band-limited, some information relating to true brightness will still be lost, even if no such bandwidth limitation is applied to the luminance signal. In this case, failure of constant luminance manifests itself as, for example, darkening of regions around chrominance transitions and loss of luminance detail in saturated chrominance areas.

The problem of failure of constant luminance has been known for many years, and various methods have been described that attempt to mitigate the effect. For example, BBC Research Department Report No. 1972/29 (published by The British Broadcasting Corporation, Research Department, Engineering Division, Kingswood Warren, Tadworth, Surrey, England, August 1972) describes the problem and proposes a method of correcting for the effect in a video coder by producing a modified Y' signal, without requiring any changes in the decoder.

The principle of the method described in that report is illustrated in Figure 1 of the drawings accompanying the present application. This figure does not appear in the report itself, but is based upon it. The circuit 50 for correcting for the failure of constant luminance receives gamma-corrected source signals R', G\ B' at inputs 52. A normal RGB to YUV matrix 54 forms Y'U'V signals from gamma-corrected R'G'B' signals in the conventional manner. The low-frequency part of the modified Y' signal (having frequencies within the passband of the chrominance channels) is formed from the low-frequency part of the conventional Y' signal by a low-pass filter (LPF) 56.

Frequencies in the modified Y' signal that are outside the passband of the chrominance channels are derived by highpass filtering the gamma-corrected true luminance signal with a filtering function which is complementary to the low-pass filter 56. This signal is formed directly from the linear red, green and blue signals using weighting factors that correctly represent the perceived brightness of the three primary phosphors. To that end, the three input signals R\ G', B' are applied to inverse gamma correction circuitry 60 to produce linear R, G, B signals. A matrix 62 combines these to produce true luminance. The linear true luminance signal from the matrix is then applied to a normal gamma correction circuit 64. The output of the gamma correction circuit 64 is then applied to two combining circuits 66,68. The circuit 66 is a subtracter with its inverting input connected to the output of the gamma correction circuit 64, its non-inverting input connected to the output of the matrix 54, and its output connected to the input of the low-pass filter 56. The circuit 68 is an adder with one input connected to the output of the gamma correction circuit 64, its other input connected to the output of the low-pass filter 56, and its output forming the modified Y' signal.

At low frequencies the filter 56 passes the low frequency Y' signal from matrix 54. The low frequency output of gamma correction circuit 64 is subtracted from the output of matrix 54 in subtracter 66 and added back to the signal in adder 68, and is thus ignored over all. At high frequencies, the filter 56 passes nothing, and the output of the gamma corrector circuit 64 is passed to the output by the adder 68. The modified Y' signal is sent to the transmission channel along with the chrominance signals U', V.

The arrangement shown is equivalent to connecting the matrix 54 directly to the filter 56, omitting the subtracter 66, and connecting the gamma correction circuit 64 to the adder 68 through a high-pass filter that is exactly complementary to the low-pass filter 56. The above reference explains the operation of the system by considering low-frequency and high-frequency signals separately. For low frequency signals (within the chrominance channel bandwidth), the correct brightness will be displayed, since no transmitted information is lost. For high frequency signals (outside the chrominance channel bandwidth but within that of the luminance channel), the reference assumes that the received chrominance signals must be zero. Therefore the decoder works in the same way as a 'monochrome' television receiver, producing a signal whose brightness is determined by the received luminance signal Y'. Since, at these frequencies, the transmitted luminance signal modified Y' has been formed in such a way as to represent true luminance, the true luminance will be displayed. We have realised that the solution proposed in the aforementioned reference does not provide a complete cure for the failure of constant luminance. The inherent assumption on which the solution is based is that a given area of the image contains chrominance and luminance frequencies that are either both high or both low, but not high-frequency components of one and low-frequency components of the other. When this assumption is invalid, the system fails to work as intended, for two reasons.

The first problem is that if the chrominance signals vary without a net change in true luminance, the system discussed above is blind to the changes (since the U' and V signals are not involved in the derivation of the correction signal). If there is significant energy Fn the chrominance signals outside the chrominance channel passband, these signals will be attenuated and this will cause a change in the true luminance value of the displayed signal. This change will not be reflected in the derived correction signal. Even in the more general case, where chrominance changes are accompanied by some change in true luminance, use of the high-pass filtered true luminance signal will not necessarily yield the true luminance level when the signal is decoded and displayed.

Conversely, the second problem is that if the true luminance of the source image varies in areas of the image containing low chrominance frequencies, the high-frequency part of the gamma-corrected true luminance signal which forms the high-frequency part of the modified Y' signal will not necessarily cause the desired change in true luminance of the decoded and displayed image. This is because of the non-linearity (the 'gamma law') of the CRT; although the high-frequency part of the modified

Y' signal is derived from the gamma-corrected true luminance, this gamma correction will only be appropriate for the non-linearity of the CRT if each of the three guns is operating at the same point on its transfer function corresponding to the amplitude of the true luminance signal (ie the image is grey). If the chrominance signals are non-zero, then the R', G' and B' signals will all be at different levels, and therefore the small-signal gains of each will be different.

An example that illustrates the first of these situations will now be considered. Figure 2 shows various waveforms in the system of Figure 1. Figure 2 shows the signal levels for two adjacent coloured areas across the image as follows: (a) the true luminance of the source and the 'conventional' transmitted luminance Y';

(b) the U and V signals directly from the source;

(c) the U' and V signals after band-limiting in the coder; and (d) the true luminance of the received image.

The area on the left is magenta and the area on the right is green. The transmitted conventional Y' signal for the areas shown at (a) is approximately the same. The 'true' luminance of the two areas is greater than the value of conventional Y', since some of the true luminance is being conveyed in the U' and V signals, since the areas are highly saturated. In this example, the degree of saturation of the two areas has been chosen to make the true luminance of the two areas approximately the same. When the YUV signals are coded, the U and V signals at (b) are band-limited, resulting in signals U' and V which exhibit softening of the transition between the two regions, as shown at (c). Thus the saturation of the image decreases near the boundary. This results in a decrease in the perceived luminance of the decoded signal when it is displayed, as shown at (d). If the amplitudes of U' and V are equal and opposite either side of the transition, then the low-pass filtered chrominance signals will pass through zero at the centre of the transition. In this case the true luminance of the decoded signal will fall to being exactly equal to the transmitted Y value. The system described in the aforementioned reference will make no change to the transmitted luminance signal in this situation; it cannot, since there are no high-frequency components in the true luminance of the source signal. Attention is drawn to US Patent No. 4,999,702 which describes an improvement to an NTSC encoding matrix which addresses the problem of failure of constant luminance. Chrominance signals are formed in a conventional manner, low-pass filtered at 1.2 MHz. As in the arrangement of Figure 1 , linear RGB signals are matrixed to provide a linear, true luminance signal. This is low-pass filtered at 4.2 MHz and passed to a gamma corrector circuit which takes the form of an arrangement of look-up tables and interpolators. These take information from the chrominance channels and endeavour to provide a corrected luminance output such that the outputted signal retains constant luminance despite the low-pass filtering undergone by the chrominance information.

In practical arrangements, the signal provided by circuits such as that in US Patent No. 4,999,702, is likely to undergo further filtering or other signal processing operations before conversion to colour component form and display. It will be clearly understood that if any of the downstream operations reduce the chrominance bandwidth further, or distort in some other way the chrominance information, the problem of failure of constant luminance is likely to reappear.

It is an object of one aspect of the present invention to provide an improved arrangement in which these difficulties with downstream chrominance processing are overcome or at least reduced.

The present invention in its various aspects is defined in the appended claims to which reference should now be made. Advantageous features are set forth in the dependant claims.

Various preferred embodiments of the invention are described in more detail below with reference to the drawings. Some of these provide ways of deriving a correction signal to add to the transmitted luminance which produces a perceived luminance level which is correct or very nearly correct when the signal is decoded and displayed. They work under circumstances including those for which the systems described above fails.

Instead of analysing the true luminance signal with a high-pass filter or its equivalent, the signal is compared to a measure of the perceived luminance produced from the signals from a 'local' decoder, which takes the same form as the 'remote' decoder. This local decoder is fed with YUV signals, where the U' and V signals have been subjected to the same processing (such as bandwidth limitation) that the real transmitted signals will undergo, but the Y' signal is unprocessed. Thus the perceived luminance derived locally is that which would be produced by a normal decoder fed with uncorrected luminance. The difference between the two perceived luminance signals is then added to the normal Y' signal before transmission.

The invention will now be described in more detail, by way of example, with reference to the drawings, in which: Figure 1 (referred to above) shows a prior art method of compensating for the failure of constant luminance when the bandwidth of the chrominance channel is limited;

Figure 2 (referred to above) shows waveforms of 'standard' and 'true' luminance and colour difference signals for a situation in which the method shown in Figure 1 fails to work;

Figure 3 shows a block diagram of a system embodying the present invention for correcting for the failure of constant luminance;

Figure 4 shows signal waveforms in similar form to Figure 2 for the system shown in Figure 3; Figure 5 shows a block diagram of the decoder circuitry for use in the system of Figure 3 simulating the transfer function of a decoder consisting of a YUV to RGB matrix and a cathode ray tube display; Figure 6 shows a block diagram of a second system embodying the present invention for correcting for the failure of constant luminance, using a small-signal model of the decoder to provide some correction for the deficiency of the circuit of Figure 3; Fiαure 7 shows a block diagram of a third system embodying the present invention for correcting for the failure of constant luminance , using a look-up table to determine the value of luminance signal that needs to be transmitted in order to achieve the desired value of perceived luminance.

Figure 8 illustrates a modification to part of the circuit of Figure 7.

Figure 9 shows a block diagram illustrating the present invention in a more general sense, and Figures 10 and 11 show block diagrams of still further systems embodying the present invention.

The invention will be described by way of example in the context of an RGB to YUV coder (or converter), although it may equally well be applied to a signal which is already in the form of a YUV signal. It is used at a point in the signal path before the bandwidth of the chrominance signals is reduced. Moreover, if the bandwidth of the chrominance signals is successively restricted at several places in the signal path, the method may be applied a number of times, respectively prior to each restriction. Figure 3 shows a block diagram of an RGB to YUV coder embodying the invention. The RGB signal is assumed to have been gamma-corrected. The R', G', B' gamma-corrected source signals are received at inputs 35. A signal 3 representing the true luminance of the image is generated by first passing the incoming RGB signals through compensating delays 37, converting the incoming RGB signal back to a linear form by a non-linear mapping circuit 1 , and then forming a weighted sum of the linear R, G and B signals in a matrix 2. A matrix which is suitable for use on signals corresponding to System I phosphors and illumiπant D_κ white point is:

Y^ = 0.2220 R + 0.7067 G + 0.0713 B The incoming RGB signal is also applied to a normal RGB to

YUV matrix 4, to generate the signals Y', U', Y that would be produced in a conventional RGB to YUV coder. The U' and V signals are low-pass filtered by filters 5, which simulate the effect of the transmission channel or coding system. The Y' signal is passed through a delay 6 to compensate for the delay in the chrominance filters 5. The delayed luminance and band-limited chrominance signals are then applied to a YUV to RGB matrix 7, which parallels or imitates the operation of the matrix in a normal decoder. The resulting RGB signals are then processed to derive a signal 10 corresponding to the perceived luminance of the image they represent, using a non-linear mapping circuit 8 which simulates the transfer function of a CRT, and a matrix circuit to form a weighted sum 9, in the same manner as the true luminance of the source signal was measured by circuit blocks 1 and 2. The circuits 7, 8 and 9 mirror the operation of a decoder in a receiver, and thus constitute a local decoder.

The signal 3 therefore represents the true luminance of the source image and the signal 10 represents the perceived luminance that would actually be displayed by a normal decoder if it was fed by a normal coder. The difference between these two signal represents the error caused by the 'failure of constant luminance'.

Figure 3 shows a simple way of deriving a correction to the transmitted Y' signal from these two signals. The two signals are gamma-corrected, by non-linear transfer function circuits 11 and 12, to convert them into signals pre-distorted for the transfer characteristic of a CRT. The difference between these two gamma-corrected signals is formed by a subtracter 13, the inverting input of which is connected to the circuit 12 and the non-inverting input of which is connected to the circuit 11 , yielding an error signal 14. This signal is then added to the conventionally-formed Y' signal by an adder 15 to form a corrected luminance signal. The conventionally-formed Y' signal, as well as the U' and V signals, are first delayed by delays 16 to compensate for the delay in the generation of the correction signal. This delay will be due primarily to the low-pass filters 5.

The circuitry of Figure 3 can be implemented either with analogue circuitry, or preferably with digital circuitry. If digital circuitry is used, the non-linear circuits can be implemented using read-only memories, and the matrix operations can be carried out with commercially-available VLSI circuits designed specifically for performing matrix operations on digital video signals. This arrangement works considerably better than the system shown in Figure 1 , since the effect on the perceived luminance signal of the band-limiting applied to the chrominance signals is measured directly. Consider, for example, the waveforms shown in Figure 4, which correspond to the example shown in Figure 2. Figure 4 shows the signal levels for two adjacent coloured areas, coloured magenta and green, as follows:

(a) the true luminance of the source and the conventional luminance signal Y before gamma correction in accordance with the present invention;

(b) the U' and V signals at the input to the local decoder; (c) the true luminance signal 10 at the output of the matrix

9 in the local decoder in Figure 3;

(d) the error or correction signal 14 from the subtracter 13 in Figure 3; and

(e) the Y signal after correction at the output of the adder 15.

The true luminance generated from the output of the local decoder, shown at (c), is equal to the true luminance of the source signal, shown at (a), in areas away from the chrominance transition. In the region of the transition the amplitude of the chrominance signals is reduced by the low-pass filter in the chrominance channel, causing a reduction in the true luminance of the received signal as shown at (c). Therefore the error signal 14, which is the difference between the true luminance of the source and that of the received signal, is zero in areas away from the colour transition. In the region of the transition it increases, reaching a maximum value at the point of the transition. The transmitted luminance signal shown at (e) therefore rises in the region of the transition, becoming approximately equal to the true luminance level at the centre of the transition. This increase in transmitted - Il ¬ luminance level in the region of the transition will result in the perceived luminance of the decoded signal being closely equal to that of the source signal for all points in the image.

Note that the luminance has been pre-corrected for a reduction in chrominance bandwidth which is not suffered in the circuit itself, but in a downstream decoder.

However, the arrangement shown in Figure 3 does not provide perfect correction for the perceived luminance at the remote decoder under all circumstances. This happens when the received chrominance signals are non-zero in areas of the image where the correction signal is being applied. In such circumstances, the gamma correction applied to the two true luminance signals 3,10 by the gamma correctors 11 and 12 will not be exactly correct to compensate for the non-linearity of the CRT. This is because the red, green and blue signals will not all have the same value, and therefore each will be at a different point on the 'gamma' transfer function. The arrangement shown in Figure 3- only provides a perfect solution if the gamma correctors 11 and 12 each reflect the actual non-linearity that the added signal will encounter.

Thus the arrangement of Figure 3 corrects for the first problem in the prior art system discussed above, but not the second.

In order to compensate for the failure of constant luminance to a higher accuracy, we have appreciated that it is possible to model the actual response of a decoder using a small-signal approximation, so that the correction signal added to the normal luminance signal may be scaled to account for the effect gain of the decoder. Figure 5 illustrates the detailed operation of the matrix 7, the non-linearity circuit 8 and the weighted sum circuit 9 which determine the true luminance output of the decoder. The requirement is to determine the change in perceived luminance ΔY_lruβ of the decoded picture by a change ΔY' to the transmitted signal. Referring to Figure 5, a change ΔY' to the Y' signal at the input to the YUV to RGB matrix 7 will cause a change to the received true luminance signal 10 given by the relationship

ΔYu_υβ = ΔY' (dr/dR'.W_R + dg/dG'.W_G + db/dB'.W_B)

where dr/dR' is the gradient of the transfer function of the

'gamma' function in circuit 8 for the R signal, relating the non-linear signal R' derived in the

YUV to RGB matrix to the linear signal R at the output of the non-linearity 8;

W_R is the weighting factor for R in the weighted sum circuit 9 that forms the true luminance signal (W_R = 0.2220 as mentioned above)

and similarly for the G and B signals.

Note that the transfer function of the YUV to RGB matrix is unity for the paths Y' to R', Y' to G' and Y' to B'. The values of the gradient dr/dR' depends on the value of R, and similarly for G and B. The small-signal gain of the decoder is therefore:

ΔY_lruβ / ΔY' = dr/dR'.W_R + dg/dG'.W_G + db/dB'.W_B

Figure 6 shows a form of a coder that includes circuitry to evaluate this quantity and scale the luminance correction signal accordingly. Much of Figure 6 is the same as Figure 3 and carries the same references and will not be described again. However, the gamma correction circuits 11 and 12 are omitted and the subtracter 13 receives directly the signals 3 and 10 from the matrix circuits 2 and 20 respectively. The output of the subtracter 13 is applied to a divider 36 where it is scaled by a scale factor. To derive this scale factor, a non-linear function circuit 18, which stores the slopes of the functions stored in circuit 17, is connected to receive the outputs of the matrix 7. A matrix 19 which is the same as the matrix 20 receives the output of the non-linear circuit 18 and applies an output 21 to the divisor input b of the divider 36, which receives the output 22 of subtracter 13 at its dividend input a.

The gradients of each of the three non-linearities in circuit 17 simulating the response of the CRT are determined by non-linearities in circuit 18. The output of these three non-linearities is fed to circuitry 19 to form a weighted sum. This circuitry is identical to the circuit 20 used to form the weighted sum of the R, G and B signals themselves, since the same weighting values appear in the equation for the small-signal gain above. The weighted sum 21 is then used to scale the linear error signal 22, by dividing the error signal 22 by the small-signal gain 21 in the divider circuit 36. Since the error signal will be multiplied by the small-signal gain in the remote decoder, the overall transfer function from linear correction signal to change in true luminance of the display should be unity (assuming the small-signal approximation is valid).

This embodiment of a coder gives more accurate results than the embodiment shown in Figure 3, although the improvement gained by this approach is smaller than that of Figure 3 over the method of Figure 1. Nevertheless, the embodiment of Figure 6 still does not give exactly the correct perceived luminance level at the remote decoder, because of the approximations made in the use of a small-signal model.

A further embodiment of a coder shown in Figure 7 allows an even more accurate account to be taken of the transfer function of the decoder. The principle of this embodiment is to pre-calculate the transmitted Y' level that is required to yield the desired true luminance level, as a function of the U' and V signals at the input to the decoder. The coder contains a large look-up table, which generates the Y' signal to be output from the coder, given the desired true luminance signal and the filtered chrominance signals. Such a look-up table may be derived by modelling the processes 7, 8, 9 in Figures 3 and 4 to obtain a table of perceived luminance levels as a function of the YUV signals at the input of the matrix 7. These data can then be re-arranged to yield the required level of the Y signal at the input of the matrix 7 as a function of the true luminance signal 10 and the filtered chrominance signals. Some interpolation of the data may be necessary in order to obtain a mapping for every value of true luminance.

Another way of calculating the contents of the look-up table is by using a recursive method to calculate successive approximations to the required value of the Y' signal, until the error in the transmitted Y value is less than a chosen value. The calculation is carried out separately for each possible combination of the look-up table inputs. The well-known Newton-

Rhapson method may be used, as shown in the following example.

Let the inputs to the look-up table be designated as:

Y'_truβww_tβ_d = gamma-corrected true luminance of source (24 in Figure 7);

^'i^_mm - U signal after simulation of channel or coding impairments (26 in Figure 7);

V's_tβwi = ^v signal after simulation of channel or coding impairments (27 in Figure 7).

Let the look-up table output be designated as:

Y'_correc-_ed ⁼ pr^β-corrβcted gamma-corrected Y signal to be transmitted (28 in Figure 7).

The steps involved in the method of this example are then as follows: Step 1 : Let Y^w,,^ = crt(Y _tmβwβ_nted) where crt(...) simulates the non-linear transfer function of the CRT, defined by example by crt(x) = Λ"⁵ for a 'gamma' of 2.35;

Step 2: Let

(^{the first} approximation to required output);

Step 3: Calculate R'G'B' values corresponding to Y_∞mcΛti, U'_flltβrθd,

Ve_ered using the standard weighting values in a normal YUV to RGB matrix;

Step 4: Calculate RGB = crt(R'G'B'), the corresponding linear values of

RGB;

Step 5: Calculate the corresponding true luminance level Y^,,^, from a weighted sum of the linear RGB values, for example: Y_tmβ_Go. = 0.222 R + 0.7067 G + 0.0713 B;

Step 6: Calculate the gradient of the CRT transfer function for each of the values R'G'B', so dR = crtgrad(R') (and similariy for dG and dB) where for example crtgrad(x) = 2.35 x^³⁵ for a 'gamma' of 2.35;

Step 7: Calculate the gradient, dtrueY, of the transfer function from linear RGB to true luminance level, from a weighted sum of dR, dG and dB using the same weights as used in calculating the true luminance level in step 5. So for example:

dtrueY = 0.222 dR + 0.7067 dG + 0.0713 dB; Step 8: Calculate a correction to the estimate of the output signal, given by:

delta - (Y^w.,,,, - Y^- ^ / dtrueY,

Step 9: Calculate the new estimate of the output, given by:

Y corrected ⁼ Y corrected ⁺ d lta',

Step 10: If the magnitude of delta is greater than the desired error (for example, 0.1% of the range of Y), go back to step 3.

It is worth noting that if the first approximation for

^was derived from the normal Y' signal from the RGB to YUV converter (29 in Figure 7), then the value of delta computed in the first iteration is equal to the scaled error signal produced by the divider .36 in the embodiment shown in Figure 6, and the resulting corrected signal Y^_nα^ is thus equal to the corrected signal produced by that embodiment. It will therefore be seen that the embodiment shown in Figure 6 resembles the first pass through this iterative procedure.

In Figure 7, a gamma-corrected true luminance signal 24 is formed by circuitry identical to circuits 1 , 2 and 11 in Figure 3. This signal 24 is fed through a compensating delay 37 into a look-up table 25, together with the low-pass filtered U' and V signals 26 and 27. These chrominance signals filtered by circuits 5 correspond to the chrominance signals at the input to the YUV to RGB matrix in a remote decoder. The look-up table 25 produces an output Y' on line 28 which has been pre-calculated to yield the true luminance level 23 when decoded and displayed as modelled by the blocks 7, 8, 9 of Rgure 3. The 'conventional' Y' signal formed in the matrix 29 corresponding to the matrix 4 is not required in this embodiment. The non-linear circuit 11 (which corrects for the CRT transfer function) could be incorporated into the look-up table 25; however the implementation of the circuitry may be easier if it is not. Since the circuit is likely to be realised with digital electronics and more bits are required to represent a linear signal 23 compared with a gamma-corrected signal 24, the look-up table 25 will be significantly smaller if the non-linear circuit 11 is implemented separately. Further measures may also be necessary to limit the total number of bits at the input to the look-up table, to ensure that the arrangement is practical. For example, the number of bits used to represent the U' and V signals can be less than that used for luminance. A possible configuration would be to use 10 bits to represent the gamma-corrected true luminance 24, 5 bits to represent V on line 27, and 4 bits to represent U on line 26. This requires a look-up table with a 19-bit address range (512K words). Such devices are readily available. in a modification, illustrated in Figure 8, the look-up table 25 receives the same input but is adapted to provide an output which is not the corrected signal YO^a_e ^{DUt tne} difference between that desired signal and the gamma corrected true luminance signal available on line 24. This true luminance signal is taken along a bypass path to an adder 81 at which it is added to the output from the look-up table 25 to provide the signal Y'-w-_βc_βd on line 28. Since the range over which the output of the look-up table 25 will vary has been reduced from that of the full luminance signal to that of a correction, less bits are required to achieve the same level of accuracy. Conversely, a higher level of accuracy can be achieved for a given address range.

In the above description it has been assumed that gamma- corrected single colour or colour component signals R' G' B' have been received. If the uncorrected signals are available, then these can be applied directly to the matrix 2 without the need for the inverse gamma corrector circuitry 1. Instead, a gamma corrector will be located before the matrix 4. Although the invention has been described here in the context of an RGB to YUV converter, it will be apparent that the principles can be applied equally well to a signal which is already in YUV form, as long as the processing is carried out prior to the filtering of the chrominance signals. Considering the embodiment of Figure 7 for example, the RGB to YUV matrix 29 would be replaced with a matrix performing the inverse function, i.e. YUV to RGB, with signals flowing from right to left. The incoming Y' U' V signals would be applied to the YUV inputs of this matrix, indicated by the arrow 30 in Figure 7. The matrix then produces corresponding RGB signals 31 , replacing the input signals in the earlier configuration. The remainder of the circuitry is unchanged.

In any event it will be appreciated that the system has R G B signals which are combined to form a true luminance signal, in matrix 2, and gamma-corrected chrominance signals U' and V which are band pass filtered to reflect the band-limiting that takes place in the coder.

It is constructive to regard the filters 5 of Figure 7 as being a replica of the filtering that takes place, more generally, downstream of an appropriate signal channel. Indeed band-pass filtering can then be seen as but one example of a form of post-channel chrominance processing. It is helpful here to refer to refer to Figure 9 which shows an input terminal 80 receiving a video signal on three lines; this video signal may be in the form of linear RGB, gamma-corrected RGB or even YUV. There is contained within input processor 82 the appropriate gamma or gamma removal circuits, and RGB or YUV matrices so as to provide linear RGB output at 84 and gamma-corrected chrominance at 86. The linear

RGB signals are taken to a true luminance matrix 88 providing a linear true luminance signal which is gamma-corrected at block 90 and passed to a luminance pre-corrector 92.

The gamma-corrected chrominance will typically pass through some form of pre-channel chrominance processing before being delivered, with the pre-corrected luminance, to a signal channel shown schematically at 96. This optional pre-channel chrominance processing 94 may take the for of a band pass-filter but this should be regarded as simply one example. Depending on the form of the signal channel, and some specific examples will be described shortly, the pre-channel chrominance processing can include compression or modulation. The signal channel 96 is assumed to deliver luminance information at 98 and chrominance information at 100, the latter chrominance information passing through post-channel processing at 102. Again, the form of post-channel processing will vary widely with the nature of the signal channel 96. Finally, Y,U and V signals are taken through an inverse matrix 104 to provide RGB inputs to a display 106, such as a CRT.

Turning more specifically to the improvement afforded by the present invention, the processing upstream of the signal channel 96 includes at 108 a replica of the post-channel chrominance processing. This replica 108 receives the chrominance information in the form in which it is outputted to the signal channel 96 and supplies information to the luminance pre-corrector 92. In the specific situation envisaged in Figure 7, above, this replica of post-channel chrominance processing comprises U and V low- pass filters. It can be understood, more generally, that the replica of post- channel chrominance processing 108 enables the luminance pre-corrector 92 to deliver to the signal channel 96 a pre-corrected luminance signal.

This is "pre-corrected" in the sense that the intention is to ensure compliance with constant luminance not in the luminance/chrominance information supplied to the signal channel but in the luminance/chrominance information downstream of the signal channel, after the post-channel processing of chrominance. This ability to pre-correct luminance is a powerful feature of the invention.

The invention will find particular application in pre-processing video signals before coding and transmission systems such as CCIR Recommendation 601 , PAL, MAC and MPEG, each of which imposes significant reductions in the bandwidths of the chrominance signals compared to that of the luminance. These systems may include operations such as sub-sampling, quantisation, and bit-rate reduction. In many cases, the effect of the reduction in chrominance bandwidth is apparent more because it changes the perceived luminance level of the displayed image than because of softening of chrominance information. The present invention can be used to restore the perceived luminance level to its original value.

The system has been explained as a means of correcting for the effect on perceived luminance of a low-pass filtering operation on the chrominance signals. However it may be applied to correct the perceived luminance for other distortions that the chrominance signals undergo as a part of a coding, recording or transmission process. For example, consider a coding system in which the chrominance signals are vertically filtered and subsampled. The system can be used to correct for the effect of the vertical filtering by choosing the filters 5 shown in Figures 3, 6, 7 to be vertical filters, having a response equal to the combined response of the pre-filters in the coder and the post-filters in the decoder. The effect of aliasing caused by the vertical subsampling operation may also be corrected for by replacing each filter with a pre-filter, a sub-sampler and a post-filter, to mimic the chrominance response of the coding system.

For a specific example of an alternative use of the present invention along these lines, reference is directed to Figure 10 which incorporates an MPEG2 signal channel. As will be understood, this assumes decimation of the chrominance information, effectively discarding alternate lines of chrominance signals. In other words, there is, in the upstream processing, a conversion from the 4:2:2 format to 4:2:0 format. Turning to the detail of Figure 10, gamma-corrected RGB signals are taken through a conventional matrix 100 delivering signals to pre-channel chrominance processing in the form of a 2:2 to 2:0 decimator 102. This delivers essentially a single chrominance signal to the signal channel which includes an MPEG2 coder 104 and an MPEG2 decoder 106. Downstream of the signal chain, post-channel chrominance processing is provided in the form of a 2:0 to 2:2 converter 108 which provides an inverse matrix 110 with U and V signals to accompany the Y signal from the MPEG 2 decoder. RGB outputs from the inverse matrix 110 drive the CRT 112.

In common with the previous embodiments, the gamma- corrected RGB signals are taken through a gamma removal circuit 114 providing linear RGB signals to a matrix 116 which delivers linear true luminance. This is optionally gamma-corrected at 118 before passing to the luminance pre-corrector. Gamma correction is "optional" in the sense that this function can alternatively be incorporated within the function of the luminance pre-corrector. In this specific embodiment a 2:0 to 2:2 converter 122 is provided as a replica of unit 108 in the downstream processing. This provides the necessary information to the luminance pre-corrector 120 to enable it to deliver a pre-corrected luminance signal which ensures constant luminance, not necessarily within the signal chain but in the YUV signals provided to the inverse matrix 110. The vertical filtering employed as pre- and post-channel filtering in the MPEG2 example will be present in other arrangements including composite coding schemes such as Weston Clean PAL and PALplus. A similar situation may arise in conversion from high definition standards. A further illustrative example of the application in the present invention will now be described with reference to Figure 11. In this application, the signal channel incorporates video tape recording. For convenience, those components which are shared in common with Rgure 10 retain the same reference numerals and will not be described in detail. As shown in Rgure 11 , chrominance information from the matrix 100 is low-pass filtered at 150 and taken to AM modulator 152. The output of luminance pre-corrector 120 is taken through a low-pass filter 154 to FM modulator 156. The outputs of the FM and AM modulation, respectively, are summed at 158 and form the input for video tape recording. The downstream processing comprises complimentary low- pass and high-pass filters 160 and 162 feeding respective demodulators 164 and 166 which provide the YUV inputs to the inverse matrix 110. Accordingly, in chrominance terms, the post-channel processing comprises low-pass filter 160 and demodulator 164. The present invention proposes the replication of this processing function in the upstream processor with the output from modulator 152 being taken through a low-pass filter 168 and demodulator 170 to provide chrominance information for the pre-corrector

120.

It is recognised that video tape recording such as VHS, imposes severe bandwidth restrictions on both luminance and chrominance. The loss of chrominance information is, however, the more severe and, therefore, the failure of constant luminance remains a real problem.

Application of the present invention will ensure that chrominance transition, whilst generally degraded in terms of saturation, will be much less degraded in terms of intensity.

Still further applications of the invention will occur to the reader.

Claims

1. Method of processing video signals to provide a luminance signal which is pre-compensated for failure of constant luminance caused by subsequent distortions in the coding or transmission of chrominance; the method comprising the steps of receiving a video signal on three lines; deriving from said video signal and outputting gamma-corrected chrominance; deriving from said video signal true linear luminance information; distorting the gamma-corrected chrominance information to imitate downstream coding or transmission; utilising said true linear luminance information and said distorted gamma-corrected chrominance information to derive a pre-compensated luminance signal which ensures compliance with constant luminance after said downstream chrominance distortion; and outputting said pre-compensated luminance signal.

2. A method of processing video signals to provide a luminance signal which is compensated for failure of constant luminance caused by distortions in the coding or transmission of the chrominance signals, the method comprising:

(a) receiving three input signals and providing therefrom at least three linear colour component signals and two non-linear chrominance signals;

(b) distorting the chrominance signals to imitate the coding or transmission distortion; and

(c) generating an output luminance signal from the three linear colour component signals and the two distorted chrominance signals, the generation operation comprising or being equivalent to the following functions:

(d) forming a first linear luminance signal from the linear colour component signals; and

(e) forming a corrected luminance signal from the first linear luminance signal and the distorted chrominance signals.

3. A method according to Claim 1 or Claim 2, in which the distortion step comprises low-pass filtering.

4. A method according to Claim 1 or Claim 2, in which the distortion step comprises vertical filtering.

5. A method according to Claim 2, in which the step (e) of forming a corrected luminance signal includes applying a non-linear function to the first linear luminance signal.

6. A method according to Claim 2, in which the providing step (a) includes providing an original non-linear luminance signal, and the step (e) of forming a corrected luminance signal comprises generating a second linear luminance signal from the original non-linear luminance signal and the two distorted chrominance signals.

7. A method according to Claim 6, in which the step of generating a second linear luminance signal comprises transcoding the original non- linear luminance signal and the two distorted chrominance signals into decoded colour component signals, applying an inverse non-linear function to the decoded colour component signals, and forming a luminance signal from the three resultant signals.

8. A method according to Claim 6, in which the step (e) of forming a corrected luminance signal includes applying a non-linear function to the first linear luminance signal, applying a non-linear function to the second linear luminance signal, and combining the difference between the two resultant signals with the original non-linear luminance signal.

9. A method according to Claim 7, in which the step (e) of forming a corrected luminance signal includes subtracting the first and second linear luminance signals to provide a difference signal, and scaling the difference signal in accordance with the slopes of the non-linear function.

10. A method according to Claim 2, in which the step (c) is executed with the use of a look-up table addressed by the first non-linear luminance signal or a signal derived therefrom and the two non-linear chrominance signals.

11. A video signal chain comprising upstream processing means for receiving a video signal on three lines and outputting gamma- corrected luminance and chrominance information to a signal channel and downstream processing means containing post-channel chrominance processing means and providing three colour component signals to a display having a gamma characteristic; wherein the upstream processing means includes a replica of said post-channel chrominance processing means, which replica operates on chrominance information as output to the signal channel, the upstream processing means further including luminance corrector means which receives true luminance information derived from said video signal and utilises information from said replica to derive for outputting to the channel a pre-corrected luminance signal which ensures compliance with constant luminance after post-channel chrominance processing in said downstream processing means.

12. Apparatus for processing video signals to provide a luminance signal which is compensated for failure of constant luminance caused by distortions in the coding or transmission of the chrominance signals; the apparatus comprising: means (1 ,2) for generating a first linear luminance signal representing the true brightness of the video signal; means (4) for generating two non-linear chrominance signals; means (5) for distorting the non-linear chrominance signals to imitate the coding or transmission distortion; means (7,8,9) for forming a second linear luminance signal from the distorted chrominance signals representing the perceived brightness of the video signal; means (11,12,13; 13,18,19,36) for comparing the first and second linear luminance signals or signals derived from them to provide a correction signal; and means for combining the correction signal with a non-linear luminance signal to provide a corrected luminance signal.

13. Apparatus according to Claim 12, in which the distorting means comprises low-pass filters (5).

14. Apparatus according to Claim 12,^' in which the comparing means comprises means (11) for applying a non-linear function to the first linear luminance signal.

15. Apparatus according to Claim 12, in which the comparing means comprises means (7,8,9) for generating a second linear luminance signal from the two distorted chrominance signals and an original non-linear luminance signal.

16. Apparatus according to Claim 15, in which the second linear luminance signal generating means comprises means (7) for transcoding the original non-linear luminance signal and the two distorted chrominance signals into decoded colour component signals, means (8) for applying an inverse non-linear function to the decoded colour component signals, and means (9) for forming a luminance signal from the three resultant signals.

17. Apparatus according to Claim 15, in which the comparing means includes means (11) for applying a non-linear function to the first linear luminance signal, means (12) for applying a non-linear function to the second linear luminance signal, and means (13,15) for combining the difference between the two resultant signals with the original non-linear luminance signal.

18. Apparatus according to Claim 16, in which the comparing means includes means (13,36) for subtracting the first and second linear luminance signals to provide a difference signal, and for scaling the difference signal in accordance with the slopes of the non-linear function.

19. Apparatus according to Claim 12, in which the comparing means and the combining means are constituted by means including a look¬ up table (25) which receives at least the first linear luminance signal or a signal derived therefrom, and the two distorted chrominance signals.